50 of 51 criteria have a sourced finding; 43 were also cross-checked by a second agent. Every comparison that includes Rovai AI links here for scope, limitations, atomic answers and sources.
Humans approve the rubric, templates and prototype; they do not review every finding. An agent source cross-check is not a hands-on product test. A missing sourced finding is not evidence that a capability is absent.
Electron Desktop (macOS arm64/x64, Windows x64), a browser workspace through Desktop Web (port 8766) or a standalone Rovai Server (macOS, Linux x64, Windows) usable from phone browsers, Feishu/Lark/DingTalk channels on Desktop, and a built-in rovai CLI for teammates. No native iOS/Android app, no Linux desktop app, no editor extension.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server), Rovai Server (rovai-server host package for macOS, Linux and Windows), Built-in rovai CLI for teammates (private IPC) and rovai-server commands · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
mobile_apps: No iOS or Android app in the release assets or guides; mobile access is a browser or a channel bot.
cli: `rovai-server` command (paths, token, run) and a built-in `rovai` CLI that teammates use through private IPC; no general user CLI to drive conversations.
editor_extensions: No editor extension in the repository or guides.
9 atomic answers
Desktop os [supported] Desktop packages for macOS Apple Silicon and Intel and Windows x64; no Linux desktop package in the release asset list. Sources: rovai-ai:rel, rovai-ai:w-compat, rovai-ai:w-install.
Mobile apps [unsupported] No iOS or Android app in the release assets or guides; mobile access is a browser or a channel bot. Sources: rovai-ai:rel, rovai-ai:w-remote.
Mobile browser [supported] Responsive browser UI with a phone-width login and conversation; guides note a phone capture is a responsive capture, not a physical device test. Sources: rovai-ai:w-desktop-web, rovai-ai:readme.
Cli [partial] `rovai-server` command (paths, token, run) and a built-in `rovai` CLI that teammates use through private IPC; no general user CLI to drive conversations. Sources: rovai-ai:w-server, rovai-ai:cli-skill, rovai-ai:security.
Task actions [supported] Start, steer (mentions, interrupt on supported agents), approve, review files changed, stop and resume from Desktop and the browser. Sources: rovai-ai:w-exec, rovai-ai:w-approvals, rovai-ai:w-recovery.
Notifications [supported] Desktop notifications for approvals, mentions, round completed, one-on-one replies and failed runs; Feishu/DingTalk deliver results to chat. Sources: rovai-ai:w-notif, rovai-ai:w-channels.
Cross device continuity [supported] The workspace lives on its host; another browser continues the same conversation and runs, but separate instances do not sync. Sources: rovai-ai:w-remote, rovai-ai:w-server-maint.
Connects already installed coding Agents through native protocols and ACP: Codex CLI, Claude Code, Pi, OpenCode, GitHub Copilot CLI, Kiro, Qoder, CodeBuddy, Qwen Code, TRAE CLI CN, Kimi Code, Grok Build, DeepSeek Harness, MiniMax Code, Kilo, ZCode and Antigravity; Cursor, Cline and Command Code are not generally available. Teammates in one conversation can use different Agents.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
model_providers: Models come from each Agent's catalog; DeepSeek Harness can add providers and llm-pi-ai models; Rovai holds no provider accounts.
custom_adapters: Contributors can add adapters (ACP transport, native adapters); no end-user custom adapter UI.
5 atomic answers
Harnesses [supported] Seventeen documented agents with host entry points; platform availability varies per adapter; Cursor awaits qualification and Cline/Command Code are deferred. Sources: rovai-ai:integrations, rovai-ai:w-agent-setup.
Model providers [partial] Models come from each Agent's catalog; DeepSeek Harness can add providers and llm-pi-ai models; Rovai holds no provider accounts. Sources: rovai-ai:w-agent-setup, rovai-ai:w-agents.
Custom adapters [partial] Contributors can add adapters (ACP transport, native adapters); no end-user custom adapter UI. Sources: rovai-ai:integrations, rovai-ai:acp.
Session integration [supported] Rovai drives each runtime through its native protocol (App Server, stream JSON, JSONL RPC, ACP) rather than launching a terminal. Sources: rovai-ai:integrations, rovai-ai:acp.
Mixed harness operation [supported] Teammates in one conversation can use different Agent products (README demo: Codex CLI, Claude Code, Pi, DeepSeek Harness). Sources: rovai-ai:readme, rovai-ai:w-members.
Reuses each Agent's own installed login and model configuration; no Rovai account, credits or key handling. Subscriptions vs API keys depend on the Agent; local models only through Agents that support them.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
api_accounts: Agents may use API keys per their own setup (e.g. DeepSeek Harness providers); Rovai stores no keys.
vendor_credits: No Rovai credits; the product is free.
local_models: Possible through Agents that support local models; none is managed by Rovai.
account_switching: Agent per teammate; multiple accounts of one Agent are not managed by Rovai.
provider_restrictions: Agent vendors' terms for use through Rovai are not discussed; model requests follow the Agent's provider.
7 atomic answers
Agent subscriptions [supported] Existing native credentials and configuration are reused; Agents are installed and signed in on the host. Sources: rovai-ai:w-agent-setup.
Api accounts [partial] Agents may use API keys per their own setup (e.g. DeepSeek Harness providers); Rovai stores no keys. Sources: rovai-ai:w-agent-setup, rovai-ai:product.
Vendor credits [unsupported] No Rovai credits; the product is free. Sources: rovai-ai:readme.
Local models [partial] Possible through Agents that support local models; none is managed by Rovai. Sources: rovai-ai:w-agent-setup.
Account switching [partial] Agent per teammate; multiple accounts of one Agent are not managed by Rovai. Sources: rovai-ai:w-agents.
Vendor account requirement [supported] No Rovai account is needed; Feishu/DingTalk need provider accounts. Sources: rovai-ai:readme, rovai-ai:w-channels.
Provider restrictions [unknown] Agent vendors' terms for use through Rovai are not discussed; model requests follow the Agent's provider. Sources: rovai-ai:w-agent-setup.
Documented path: install Desktop (or Server), install and sign in to one Agent, check Settings → Agents, create a teammate, open a project conversation and send a small task. Works with Git or non-Git directories; Git projects get managed Mission worktrees; no hosting account needed. Server installer and remote access recipes are documented.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
hosted_git: No GitHub/GitLab integration; delivery is by the Agent's own git commands.
managed_cloud_setup: No managed cloud.
7 atomic answers
First task setup [supported] Quick start: prepare one Agent, choose a teammate, give a small task, inspect the run. Sources: rovai-ai:w-quick, rovai-ai:readme.
No git [supported] For a non-Git directory execution retains the original directory with no automatic branch isolation. Sources: rovai-ai:w-missions.
Local git without remote [supported] Mission worktrees are created from the local HEAD with no remote involved. Sources: rovai-ai:w-missions.
Hosted git [partial] No GitHub/GitLab integration; delivery is by the Agent's own git commands. Sources: rovai-ai:srch-pr.
Prerequisite guidance [supported] Install pages list OS packages, Visual C++ runtime, glibc baseline, ordinary user account and Agent install links. Sources: rovai-ai:w-server, rovai-ai:w-agent-setup.
Managed cloud setup [unsupported] No managed cloud. Sources: rovai-ai:srch-cloud.
Byo vps onboarding [supported] Server install guide with installer script, checksums, data directory, Tailscale and reverse-proxy recipes for a user-owned host. Sources: rovai-ai:w-server, rovai-ai:w-remote.
Teammates use a built-in authenticated `rovai` CLI over private IPC; the Server has a browser/API surface for its own UI with token login; users have `rovai-server` management commands. No public REST API, SDK, MCP server or webhooks; an open issue requests a headless dispatch interface.
Scope
Surfaces: Built-in rovai CLI for teammates (private IPC) and rovai-server commands, Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
cli: Built-in `rovai` CLI for teammates (messages, members, missions, history, memory); `rovai-server` has paths/token/run commands; not a general user CLI.
api: No public documented API; issue 1017 asks for an official headless dispatch interface.
sdk: No SDK.
mcp_server: Rovai configures MCP servers for Agents; bounded documentation and source search found no MCP server exposing Rovai to other agents (the rovai CLI is the interface).
webhooks: No webhooks found.
operations: Agents can send messages, read thread state and Missions and update their profile via the rovai CLI; no external create/steer/stop API.
stability: No stability statement; contract documents carry version numbers (e.g. Camp Message Send v31) and release cadence is every few days.
8 atomic answers
Cli [partial] Built-in `rovai` CLI for teammates (messages, members, missions, history, memory); `rovai-server` has paths/token/run commands; not a general user CLI. Sources: rovai-ai:cli-skill, rovai-ai:w-server-maint.
Api [unsupported] No public documented API; issue 1017 asks for an official headless dispatch interface. Sources: rovai-ai:readme, rovai-ai:security.
Mcp server [unsupported] Rovai configures MCP servers for Agents; bounded documentation and source search found no MCP server exposing Rovai to other agents (the rovai CLI is the interface). Sources: rovai-ai:w-mcp, rovai-ai:cli-skill.
Operations [partial] Agents can send messages, read thread state and Missions and update their profile via the rovai CLI; no external create/steer/stop API. Sources: rovai-ai:cli-skill.
Authentication [supported] Server and Desktop Web use a single-owner login token; the rovai CLI authenticates over private IPC (authorization bypass is a named security scope). Sources: rovai-ai:w-server-maint, rovai-ai:security.
Stability [unknown] No stability statement; contract documents carry version numbers (e.g. Camp Message Send v31) and release cadence is every few days. Sources: rovai-ai:a2a.
MCP servers are configured in Settings → MCP (add JSON or import) and assigned to teammates, subject to each Agent's support; native Agent Skills are inventoried and selectable with `/`; Rovai Toolbox collaboration instructions are assigned per teammate. Credentials follow each server's mechanism.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
plugins: No plugin system; issue 1096 requests plugin support.
custom_tools: MCP servers and native Skills extend tools; no custom tool SDK.
authentication: Credentials use each MCP server's mechanism; Rovai keeps secrets out of messages.
permission_scope: Per-teammate server assignment; Agent permissions apply.
7 atomic answers
Mcp client [supported] Settings → MCP adds or imports server configuration and assigns teammates; availability depends on the Agent. Sources: rovai-ai:w-mcp.
Custom tools [partial] MCP servers and native Skills extend tools; no custom tool SDK. Sources: rovai-ai:w-mcp, rovai-ai:w-skills.
Instructions import [supported] Native Agent Skills (user and project scope) can be selected; MCP configs import from this computer. Sources: rovai-ai:w-skills, rovai-ai:w-mcp.
Authentication [partial] Credentials use each MCP server's mechanism; Rovai keeps secrets out of messages. Sources: rovai-ai:w-mcp.
Desktop is Electron with a Rust host (`rovai-host`) and a React renderer shared with the browser UI; Server runs the same host without Electron.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
4 atomic answers
Desktop implementation [supported] Electron application (PRODUCT.md) with Rust core and host crates. Sources: rovai-ai:product, rovai-ai:readme.
Mobile implementation [supported] No mobile app; phone use is the responsive web UI. Sources: rovai-ai:w-remote.
Hybrids [supported] Desktop and Server share the Host implementation; Desktop adds channels. Sources: rovai-ai:readme.
Implementation evidence [supported] README architecture overview, PRODUCT.md and the repository layout (apps/desktop, apps/web, crates/rovai-host). Sources: rovai-ai:readme, rovai-ai:product.
Intake is a message in a project conversation, a Mission goal, a scheduled saved prompt, or a Feishu/Lark/DingTalk message; no external issue import or Git-event triggers.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
external_import: No issue import in the pinned source or guides.
chat_git_triggers: Feishu/Lark/DingTalk messages reach teammates on Desktop; no Git triggers.
stable_ids: Threads, Missions, Tasks and messages have internal IDs but no tracker keys.
sync: No tracker sync.
supported_edits: No external records to edit.
9 atomic answers
Local creation [supported] Create conversations, Missions, Tasks and automations in the app. Sources: rovai-ai:w-missions, rovai-ai:w-tasks.
Missions hold a goal and boundaries; Tasks hold scope, requirements and expected result with status; conversations keep requests; no plan editor or acceptance-criteria object beyond text.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
plans: Mission goal and Task description act as plans; no separate plan object.
acceptance_criteria: Written in the Task or Mission description; no structured field.
revisions: Memory has version history; Task/Mission edits keep no documented revision list.
Mission → conversation → Tasks → Runs is a fixed hierarchy; Tasks have one lead and no dependency graph; durable gather barriers coordinate parallel requests; a conversation can become a Mission only by creating one.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
epics: Missions play the epic role with one conversation each.
hierarchy: Fixed Mission/Task/Run levels; no arbitrary depth.
blockers: Tasks can be Blocked with a reason; no dependency links.
readiness: No dependency-derived readiness.
enforcement: No dependency enforcement; Task creation does not start work.
state_propagation: Finishing a run, completing a mission and merging are separate actions.
task_to_epic: No conversion; a Mission is created separately.
identity_retention: No growth path that retains identity.
9 atomic answers
Epics [partial] Missions play the epic role with one conversation each. Sources: rovai-ai:w-missions.
Collaborative memory (shared, teammate, between teammates) with scopes, retrieval keys, revisions, review of agent-proposed candidates and a Memory page; teammate profiles with principles; native and Toolbox skills; frozen context manifests per run. Memory cannot currently be switched off application-wide (open issue).
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
project_docs: Files preview beside conversations; no project-doc indexing.
deletion: Entries can be retired or replaced; deletion semantics are described as continue, retire or replace.
freshness: Stale candidates are flagged when their target or revision changed; no age-based staleness.
10 atomic answers
Instructions [supported] Teammate role, responsibilities and working principles; Agents read their own project instruction files. Sources: rovai-ai:w-members.
Memory [supported] Shared, teammate and relationship memory with candidate review. Sources: rovai-ai:w-memory.
Scope [supported] Shared, one teammate or a pair (mutual or directed). Sources: rovai-ai:w-memory.
Provenance [supported] Entries show formation source (user created or candidate) and version history. Sources: rovai-ai:w-memory.
Correction [supported] Revise body and keys; candidates can be edited then accepted. Sources: rovai-ai:w-memory.
Deletion [partial] Entries can be retired or replaced; deletion semantics are described as continue, retire or replace. Sources: rovai-ai:w-memory.
Freshness [partial] Stale candidates are flagged when their target or revision changed; no age-based staleness. Sources: rovai-ai:w-memory.
Takeover continuity [supported] Conversations, teammates, memory and run records live in the instance and persist across sessions. Sources: rovai-ai:product, rovai-ai:w-remote.
No code index or retrieval of its own; Agents search code. Rovai provides file preview, per-run and cumulative diffs, attachments with quotes and a frozen ContextManifest per run.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
Not separately cross-checked
code_search: No Rovai code search.
indexing: No repository code index found in the documentation or the checked source.
external_knowledge: MCP servers can reach external knowledge through the Agent.
multimodal_inputs: Attachments and runtime images are supported in messages.
freshness: Context manifests are frozen at run start; file views refresh on demand.
access_scope: Run workspace access is read_only or write with isolation shared or git_worktree.
api_retrieval: No retrieval API.
citations: Message quotes and evidence references are retained.
Native Missions (board with statuses), Tasks (owner, status, description) and Memory; no idea backlog, labels or priorities; conversations serve as discussion; external tracker not required for these but there is no import.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
prioritize: No priority field documented.
label: Missions accept labels and source attachments "when useful".
search: Jump to chat search; cross-conversation message search is an open request.
filter: Board status columns; no filter DSL documented.
archive: Cleanup and completion are explicit; conversation archive is an open request.
reopen: Mission business status can be changed; terminal Tasks stay terminal and need a follow-up Task.
idea_parking: A Mission can be created without starting (New does not start work).
Assign [supported] A Task has a lead/owner chosen from conversation members. Sources: rovai-ai:w-tasks.
Search [partial] Jump to chat search; cross-conversation message search is an open request. Sources: rovai-ai:readme.
Filter [partial] Board status columns; no filter DSL documented. Sources: rovai-ai:w-missions.
Archive [partial] Cleanup and completion are explicit; conversation archive is an open request. Sources: rovai-ai:w-missions.
Reopen [partial] Mission business status can be changed; terminal Tasks stay terminal and need a follow-up Task. Sources: rovai-ai:w-tasks, rovai-ai:w-recovery.
Project epic views [supported] Mission board per project and team. Sources: rovai-ai:w-missions.
Idea parking [partial] A Mission can be created without starting (New does not start work). Sources: rovai-ai:w-missions.
Explicit dispatch [supported] Start Mission requests execution separately; a Task does not start an Agent until addressed. Sources: rovai-ai:w-missions, rovai-ai:w-tasks.
Task session separation [supported] Mission, conversation, Task and Run are separate entities. Sources: rovai-ai:w-missions.
Durable states [supported] Tasks Pending/In progress/Blocked/Complete/Cancelled and Mission status persist in SQLite. Sources: rovai-ai:w-tasks, rovai-ai:product.
External tracker requirement [partial] Not required for Rovai's own work tracking; no tracker integration exists. Sources: rovai-ai:w-missions.
Teammates read and write the same durable Thread/Mission/memory records humans use through the built-in rovai CLI (messages, missions, history, memory, own profile) with attributed messages and recovery guidance; Task creation by agents is not documented.
Scope
Surfaces: Built-in rovai CLI for teammates (private IPC) and rovai-server commands · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
create: Send messages and (via member-studio) propose members; Mission/Task creation by agents is not documented.
update: Update Mission status/explanation and own profile; shared memory goes through candidates.
claim: A run claims deliveries from a FIFO; no task claim by agents.
assign: Agents address teammates in messages; Task assignment by agents is not documented.
state_changes: Mission status updates; Task updates by assignees for their own status.
dependencies: No dependencies.
children: Parallel requests to several members with repeated --to.
permissions: IPC authorization is a named security scope; Task permissions limit non-lead assignees.
14 atomic answers
Shared ids [supported] The rovai CLI and UI use the same thread, message and Mission IDs. Sources: rovai-ai:cli-skill, rovai-ai:a2a.
Create [partial] Send messages and (via member-studio) propose members; Mission/Task creation by agents is not documented. Sources: rovai-ai:cli-skill.
Update [partial] Update Mission status/explanation and own profile; shared memory goes through candidates. Sources: rovai-ai:cli-skill, rovai-ai:w-memory.
Claim [partial] A run claims deliveries from a FIFO; no task claim by agents. Sources: rovai-ai:a2a.
Assign [partial] Agents address teammates in messages; Task assignment by agents is not documented. Sources: rovai-ai:cli-skill.
State changes [partial] Mission status updates; Task updates by assignees for their own status. Sources: rovai-ai:cli-skill, rovai-ai:w-tasks.
Dependencies [unsupported] No dependencies. Sources: rovai-ai:w-tasks.
Children [partial] Parallel requests to several members with repeated --to. Sources: rovai-ai:cli-skill.
Results [supported] Public answers and Mission explanations are published as ThreadMessages. Sources: rovai-ai:cli-skill.
Human visibility [supported] Everything appears in the shared conversation and Run panel. Sources: rovai-ai:w-collab, rovai-ai:w-exec.
No living product specification; Mission goals, Tasks and Memory agreements are the closest records and are not requirement stores.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
Not separately cross-checked
persistent_requirements: Shared memory agreements persist across tasks but are not requirements.
agent_edits: Agents propose shared candidates for review and write their own teammate memory.
implementation_links: No requirement-to-implementation links.
later_retrieval: Teammates look up memory when asked; presence does not prove a run read it.
drift_detection: No drift detection.
review_integration: No spec-aware review.
9 atomic answers
Persistent requirements [partial] Shared memory agreements persist across tasks but are not requirements. Sources: rovai-ai:w-memory.
Human edits [supported] Memory entries can be revised by the user. Sources: rovai-ai:w-memory.
Agent edits [partial] Agents propose shared candidates for review and write their own teammate memory. Sources: rovai-ai:w-memory.
Revisions [supported] Version history on memory entries. Sources: rovai-ai:w-memory.
Approvals [supported] Pending candidates need accept, edit or reject. Sources: rovai-ai:w-memory.
Different teammates run concurrently in one conversation or across projects (FIFO per teammate); Run panel and sidebar show queued, running, waiting and finished work. No numeric quota documented.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
Not separately cross-checked
quotas: No concurrency limit is documented in the guides or architecture notes read.
observed_concurrency: No concurrency measurements published.
5 atomic answers
Concurrent sessions [supported] Each teammate's run executes independently; mentions can fan out to several teammates. Sources: rovai-ai:a2a, rovai-ai:cli-skill.
Concurrent tasks [supported] Several runs across teammates and conversations. Sources: rovai-ai:w-exec.
Quotas [partial] No cap on concurrent runs found: reusable resident agent-runtime processes are capped (20 per teammate bucket, 200 global, 30-minute idle TTL) and runs beyond that start as non-resident burst processes; agent-to-agent dispatch stops at chain depth 5 or 16 runs. Sources: rovai-ai:rev-fleet, rovai-ai:rev-delegation-cap, rovai-ai:a2a.
Observed concurrency [unknown] No concurrency measurements published. Sources: rovai-ai:readme.
Status visibility [supported] Run states queued, running, waiting for approval/input, failed, completed per teammate. Sources: rovai-ai:w-exec.
Humans address teammates by structured mention; the Default Lead receives by default; deliveries wait in per-teammate FIFO queues with claims; Tasks record an owner; writers on the same file are the user's responsibility. Claim is enforced by the scheduler.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
routing: Explicit targets; Default Lead receives unaddressed messages.
duplicate_prevention: FIFO claims prevent a second run for the same delivery; concurrent writers on one file are not prevented.
ownership_enforcement: Task updates are permission-limited; file writes are not exclusive.
Agent assignment [supported] Agents address teammates via the rovai CLI; Default Lead can request independent work from several members. Sources: rovai-ai:cli-skill.
Claims [supported] The scheduler claims one FIFO prefix of deliveries into an AgentRun. Sources: rovai-ai:a2a.
Queues [supported] Waiting deliveries and queued runs per teammate; private chat FIFO. Sources: rovai-ai:a2a, rovai-ai:w-queue.
Duplicate prevention [partial] FIFO claims prevent a second run for the same delivery; concurrent writers on one file are not prevented. Sources: rovai-ai:a2a, rovai-ai:w-collab.
Ownership enforcement [partial] Task updates are permission-limited; file writes are not exclusive. Sources: rovai-ai:w-tasks, rovai-ai:w-collab.
Product-owned team: a Default Lead and teammates exchange messages and hand off work through the built-in rovai CLI; each teammate picks its own Agent, model and permissions; runs and replies are inspectable and retained. Coordination is orchestrated by Rovai Core, with Agents doing the work.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Built-in rovai CLI for teammates (private IPC) and rovai-server commands · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Model harness effort [supported] Agent, model, reasoning parameters, filesystem access and approval policy per teammate. Sources: rovai-ai:w-agents.
Messages [supported] Public ThreadMessages with explicit targets and one-on-one DMs. Sources: rovai-ai:a2a, rovai-ai:w-collab.
Results return [supported] Replies return as separate messages; durable gather barriers collect parallel results. Sources: rovai-ai:cli-skill, rovai-ai:a2a.
Cross harness [supported] A team can mix Codex CLI, Claude Code, Pi and DeepSeek Harness. Sources: rovai-ai:readme.
Orchestration owner [supported] Product orchestration by Rovai Core; not a provider subagent feature. Sources: rovai-ai:a2a, rovai-ai:product.
Activation [supported] Available by default; agent-to-agent handoff is a supported path. Sources: rovai-ai:w-collab.
Authentication [supported] Each Agent uses its own installed login; the CLI uses private IPC authorization. Sources: rovai-ai:w-agent-setup, rovai-ai:security.
Live progress [supported] Run panel shows each teammate's steps live. Sources: rovai-ai:w-exec.
Scheduled tasks (Automations) run a saved request with one teammate on the Desktop host, each occurrence in a new conversation, with run-once, history, overlap skipping and an optional channel notification; Feishu/Lark/DingTalk messages are an event trigger on Desktop. Runs stop when the host is closed or asleep and missed occurrences are not replayed.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
events: Channel messages (Feishu, Lark, DingTalk) trigger work on Desktop; no webhook or Git events.
completion_conditions: A run ends by its own completion; approval or input needs cancel the unattended run with an interaction-required failure.
retries: Missed occurrences record the latest miss and advance; no replay or retry of failed runs.
client_independence: No browser needed, but the Desktop or Server process and its host must stay awake.
8 atomic answers
Schedules [supported] Automations → New with a schedule in the host's local time zone. Sources: rovai-ai:w-auto.
Events [partial] Channel messages (Feishu, Lark, DingTalk) trigger work on Desktop; no webhook or Git events. Sources: rovai-ai:w-channels, rovai-ai:srch-webhook.
Recurring tasks [supported] Recurring requests with execution history. Sources: rovai-ai:w-auto.
Completion conditions [partial] A run ends by its own completion; approval or input needs cancel the unattended run with an interaction-required failure. Sources: rovai-ai:w-auto.
Scheduler location [supported] Runs on the Rovai host process; not a cloud service. Sources: rovai-ai:w-auto.
Overlaps [supported] An overlapping occurrence of the same automation is skipped, not queued. Sources: rovai-ai:w-auto.
Retries [partial] Missed occurrences record the latest miss and advance; no replay or retry of failed runs. Sources: rovai-ai:w-auto.
Client independence [partial] No browser needed, but the Desktop or Server process and its host must stay awake. Sources: rovai-ai:w-auto, rovai-ai:w-server-maint.
Stop a run, withdraw an unclaimed message, interrupt/steer supported Agents mid-run, answer approvals and questions in a pending area, resume by reopening the conversation; notifications for approvals, mentions and failures; recovery guidance for uncertain outcomes.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
pause: No pause control; stop and queue are separate.
resume: Send a bounded follow-up after reading the last run state; sessions can continue on supported Agents.
steer: Mid-run instructions are supported for Agents that allow it (e.g. OpenCode 2.2.7+), otherwise a queued follow-up.
escalation: Notifications and channel messages; no escalation policy.
chat_terminal_parity: Agents are driven through protocols, not terminals; CLI-only prompts surface as approvals or waiting-for-input where the adapter supports them.
10 atomic answers
Pause [unsupported] No pause control; stop and queue are separate. Sources: rovai-ai:w-recovery.
Stop [supported] Stop the selected Run; history is retained and file writes are not rolled back. Sources: rovai-ai:w-recovery.
Resume [partial] Send a bounded follow-up after reading the last run state; sessions can continue on supported Agents. Sources: rovai-ai:w-recovery, rovai-ai:rel.
Steer [partial] Mid-run instructions are supported for Agents that allow it (e.g. OpenCode 2.2.7+), otherwise a queued follow-up. Sources: rovai-ai:rel, rovai-ai:w-queue.
Pending decisions [supported] Pending approvals dock above the composer with previous/next controls; waiting-for-input runs show the question. Sources: rovai-ai:w-approvals, rovai-ai:w-exec.
Visible errors [supported] Failed or interrupted runs show last successful step and error. Sources: rovai-ai:w-exec, rovai-ai:w-recovery.
Chat terminal parity [partial] Agents are driven through protocols, not terminals; CLI-only prompts surface as approvals or waiting-for-input where the adapter supports them. Sources: rovai-ai:w-approvals, rovai-ai:w-exec.
Mission board with statuses, sidebar project/conversation lists with unread/running indicators, a Run overview across teammates, and a Team panel; no dependency or workload view.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
parent_child: Mission → conversation → Tasks → Runs, shown in the board drawer and run overview.
dependencies: No dependency graph.
workload: Run overview shows each teammate's queue and running work.
navigation_errors: Failed/blocked states appear at run and Task level.
8 atomic answers
Fleet fixture [supported] Mission board and sidebar list conversations per project with running indicators; README screenshots use English demo teams. Sources: rovai-ai:w-missions, rovai-ai:readme.
Parent child [partial] Mission → conversation → Tasks → Runs, shown in the board drawer and run overview. Sources: rovai-ai:w-missions.
Dependencies [unsupported] No dependency graph. Sources: rovai-ai:w-tasks.
No mandatory workflow: one teammate, parallel mentions, review handoffs, Missions, tasks and schedules; per-teammate Agent/model/permissions; configuration through UI; member-studio skill can propose teammates; live roles visible in Run/Team panels.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
growing_epic: A project conversation can be promoted to work with a Mission only by creating one.
presets: Default teammates and lead for new conversations; teammate profiles.
natural_language: `member-studio` skill proposes teammates; settings are forms.
in_run_changes: Configuration changes apply to future runs; an existing run keeps frozen configuration.
auth_friction: Each Agent must be installed and signed in separately on the host.
Work runs on the machine hosting Desktop or Server (the user's computer or a user-managed host); there is no managed cloud. Server runs independently, accepted requests continue after the browser disconnects and another device can reconnect; Desktop Web needs Desktop running; the host must stay awake.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
managed_cloud: No vendor cloud execution found.
hybrid: A Desktop can be reached over Desktop Web while work stays on that host; Desktop and Server are separate instances.
provisioning: User installs the package with a script and checksums; no provisioning service.
Byo remote [supported] Rovai Server on the user's own host (macOS, Linux x64, Windows) with LAN, Tailscale or HTTPS access. Sources: rovai-ai:w-server, rovai-ai:w-remote.
Self hosted [supported] Standalone Server is a self-hosted control plane with its own data directory. Sources: rovai-ai:w-server.
Managed cloud [unsupported] No vendor cloud execution found. Sources: rovai-ai:srch-cloud.
Hybrid [partial] A Desktop can be reached over Desktop Web while work stays on that host; Desktop and Server are separate instances. Sources: rovai-ai:w-remote.
Provisioning [partial] User installs the package with a script and checksums; no provisioning service. Sources: rovai-ai:w-server.
Payer [supported] User pays their host and Agent providers; Rovai is free. Sources: rovai-ai:readme.
Resources [supported] The host's resources; no vendor limits. Sources: rovai-ai:w-server.
Region [supported] Chosen by the user's host. Sources: rovai-ai:w-server.
Expiry [supported] No vendor expiry; Server runs until stopped (foreground by default). Sources: rovai-ai:w-server-maint.
Client disconnect [supported] A closed tab or lost Wi-Fi does not stop an accepted request. Sources: rovai-ai:w-server-maint.
Other device continuation [supported] Another browser logs in to the same instance and continues the conversation. Sources: rovai-ai:w-desktop-web, rovai-ai:w-remote.
Runs use a shared project directory or, for Git Missions, a managed Git worktree shared by the Mission's participants; access is read_only or write per teammate; no container, VM or OS sandbox of Rovai's own (Agents apply their own permission modes); highest permission mode is user-selectable.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
branch: Mission worktrees are based on current HEAD; manual branch switches do not rewrite the base; branch creation is not a documented product action.
container: No container launcher in the pinned source.
vm: No VM isolation.
os_sandbox: Sandbox behavior comes from each Agent (filesystem and approval options); Rovai adds none.
filesystem: Per-teammate read-only or workspace-write filesystem access where the Agent offers it; path traversal and symlink escape are named security scopes.
process: Agents run as host processes under the user account.
network: No Rovai network controls.
secrets: Credentials stay in the Agent's native configuration; secrets are kept out of messages and evidence.
Branch [partial] Mission worktrees are based on current HEAD; manual branch switches do not rewrite the base; branch creation is not a documented product action. Sources: rovai-ai:w-missions.
Os sandbox [partial] Sandbox behavior comes from each Agent (filesystem and approval options); Rovai adds none. Sources: rovai-ai:w-agents, rovai-ai:srch-isolation.
Filesystem [partial] Per-teammate read-only or workspace-write filesystem access where the Agent offers it; path traversal and symlink escape are named security scopes. Sources: rovai-ai:w-agents, rovai-ai:security.
Process [unsupported] Agents run as host processes under the user account. Sources: rovai-ai:srch-isolation.
Secrets [partial] Credentials stay in the Agent's native configuration; secrets are kept out of messages and evidence. Sources: rovai-ai:product, rovai-ai:security.
Code separation [partial] Mission worktrees separate Mission work from the source checkout; participants within a Mission share one worktree. Sources: rovai-ai:w-missions.
Security boundary [unsupported] No hostile-code containment claimed. Sources: rovai-ai:security.
Accepted requests continue after browser disconnects; app or host closure requires reopening and inspecting run state before resending; stop is not undo; cleanup and recovery are explicit with uncertain-outcome handling; stored facts persist in SQLite; scheduled runs after sleep skip missed occurrences.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
Not separately cross-checked
worker_crash: Agent or network failure leaves a failed run with last successful step; retry is a manual bounded follow-up.
service_restart: State persists in the data directory; Desktop Web must be re-enabled after restart; runs interrupted by host shutdown are inspected before resuming.
process_reconnect: Reconnect to the same instance and inspect the existing execution; agent session reattach depends on the adapter.
task_resume: Later runs can continue the original session on supported Agents (e.g. OpenCode).
context_loss: Runs freeze context manifests; Agent-side context after a crash depends on the runtime.
manual_repair: Users inspect files and send bounded follow-ups; no rotate-token command exists for exposed credentials.
12 atomic answers
Client disconnect [supported] A closed tab does not stop accepted work. Sources: rovai-ai:w-server-maint.
Worker crash [partial] Agent or network failure leaves a failed run with last successful step; retry is a manual bounded follow-up. Sources: rovai-ai:w-recovery.
Service restart [partial] State persists in the data directory; Desktop Web must be re-enabled after restart; runs interrupted by host shutdown are inspected before resuming. Sources: rovai-ai:w-server-maint, rovai-ai:w-recovery.
Expiry [partial] No sandbox expiry; idle reusable agent-runtime processes are released after a 30-minute idle TTL (state persists in SQLite and later runs start a new process). Sources: rovai-ai:w-server, rovai-ai:rev-fleet.
Transcript reload [supported] Conversations and run steps reload from stored facts. Sources: rovai-ai:product, rovai-ai:w-exec.
Process reconnect [partial] Reconnect to the same instance and inspect the existing execution; agent session reattach depends on the adapter. Sources: rovai-ai:w-server-maint.
Task resume [partial] Later runs can continue the original session on supported Agents (e.g. OpenCode). Sources: rovai-ai:rel.
State reconstruction [supported] SQLite-backed Core facts are authoritative; renderer pages are projections. Sources: rovai-ai:product.
Context loss [partial] Runs freeze context manifests; Agent-side context after a crash depends on the runtime. Sources: rovai-ai:a2a.
Manual repair [partial] Users inspect files and send bounded follow-ups; no rotate-token command exists for exposed credentials. Sources: rovai-ai:w-recovery, rovai-ai:w-server-maint.
Multi day continuity [supported] Missions, conversations and memory persist across days; Server retains data in its data directory. Sources: rovai-ai:w-missions, rovai-ai:w-server.
No dependency, cache, service or port management; the user installs Agents and prepares project directories; Mission worktrees are created from HEAD without copying uncommitted edits; Server installer checks archives.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
dependencies: No project dependency installation.
caches: No cache handling.
snapshots: No environment snapshots.
services: No managed services.
ports: Only Rovai's own ports (8766 Desktop Web, 8767 Server) are configurable.
secrets: Agent credentials are the Agent's own; no injection.
duplicate_installs: Not addressed.
collisions: Not addressed.
10 atomic answers
Dependencies [unsupported] No project dependency installation. Sources: rovai-ai:srch-isolation.
No product time, turn, token or spend limits; scheduled occurrences skip overlaps and cancel when interaction is required; version queries time out after 10 seconds; usage is reported, not limited.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
Not separately cross-checked
time_limits: No run time limits documented.
turn_limits: No turn limits.
token_limits: No token limits; Agent-specific.
spend_limits: No spend limits; usage is informational.
timeouts: Internal timeouts (e.g. agent version query 10s); unattended runs that need approval are cancelled.
retries: No automatic retry; follow-ups are manual.
provider_failures: Failures are shown with last step and error; Agent status reports unavailable configurations.
enforcement: Only the unattended-approval cancel and overlap skip are enforced.
Turn limits [partial] No per-run turn limit documented; agent-to-agent delegation is capped per chain (new dispatch refused at depth 5 or 16 runs). Sources: rovai-ai:w-exec, rovai-ai:rev-delegation-cap.
Timeouts [partial] Internal timeouts (e.g. agent version query 10s); unattended runs that need approval are cancelled. Sources: rovai-ai:rel, rovai-ai:w-auto.
Retries [unsupported] No automatic retry; follow-ups are manual. Sources: rovai-ai:w-recovery.
Provider failures [partial] Failures are shown with last step and error; Agent status reports unavailable configurations. Sources: rovai-ai:w-exec, rovai-ai:w-agents.
Enforcement [partial] Only the unattended-approval cancel and overlap skip are enforced. Sources: rovai-ai:w-auto.
Run records with steps and states, evidence-qualified usage monitoring (tokens, cache, attributable cost) by range, agent, provider and model with JSON export, per-run native usage and context, diagnostics center and notifications; unknown values never shown as zero.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
traces: Observed tool steps and file changes; no distributed trace.
task_stats: Usage grouped by run or session; Task-level aggregation not documented.
project_stats: No project-level statistics documented.
performance_method: Evaluation and benchmark protocol documents exist but the catalog does not run them and Rovai is not on ADEMark.
responsiveness: No responsiveness measurements published.
13 atomic answers
Logs [supported] Execution steps, tool results and diagnostics per run. Sources: rovai-ai:w-exec.
Traces [partial] Observed tool steps and file changes; no distributed trace. Sources: rovai-ai:w-exec.
Failure reasons [supported] Failed or interrupted runs show error and last successful step. Sources: rovai-ai:w-exec, rovai-ai:w-recovery.
Usage export [supported] JSON export of reported raw entries. Sources: rovai-ai:w-usage.
Task stats [partial] Usage grouped by run or session; Task-level aggregation not documented. Sources: rovai-ai:w-usage.
Agent stats [supported] Filter by Agent, provider and model. Sources: rovai-ai:w-usage.
Project stats [unsupported] No project-level statistics documented. Sources: rovai-ai:w-usage.
Time stats [supported] Past 24 hours, 7 days and 30 days. Sources: rovai-ai:w-usage.
Failure status [supported] Failed, interrupted, waiting states per run. Sources: rovai-ai:w-exec.
Approval status [supported] Waiting for approval is a distinct state with the exact request. Sources: rovai-ai:w-exec, rovai-ai:w-approvals.
Reconnect consistency [supported] SQLite facts are authoritative; renderers are read-only projections. Sources: rovai-ai:product.
Performance method [unknown] Evaluation and benchmark protocol documents exist but the catalog does not run them and Rovai is not on ADEMark. Sources: rovai-ai:monitoring.
Responsiveness [unknown] No responsiveness measurements published. Sources: rovai-ai:readme.
Usage monitoring shows Agent-reported tokens, cache and attributable cost (and a public-price estimate) by call, turn, run or session; subscription marginal cost is not computed; missing data stays unknown, never zero; no compute cost.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
task_cost: Grouped by run or session; Task-level not documented.
child_cost: Per teammate run by Agent filter; no parent rollup.
epic_cost: No Mission-level cost documented.
harness_aggregation: Table spans Agents with provider and model filters.
subscription_allowance: No allowance tracking; unknown stays unknown.
provider_reconciliation: Estimates may differ from the provider bill; no reconciliation.
compute_cost: No compute fees.
alerts: No budget alerts.
14 atomic answers
Task cost [partial] Grouped by run or session; Task-level not documented. Sources: rovai-ai:w-usage.
Session cost [supported] Cost grouping by session. Sources: rovai-ai:w-usage.
Child cost [partial] Per teammate run by Agent filter; no parent rollup. Sources: rovai-ai:w-usage.
Epic cost [unsupported] No Mission-level cost documented. Sources: rovai-ai:w-usage.
Harness aggregation [partial] Table spans Agents with provider and model filters. Sources: rovai-ai:w-usage.
Billed vs estimated [supported] Runtime-reported cost, public-price estimates and billing buckets remain distinct grains. Sources: rovai-ai:product, rovai-ai:monitoring.
Tokens [supported] Token and cache usage reported by the Agent. Sources: rovai-ai:w-usage.
Provider reconciliation [partial] Estimates may differ from the provider bill; no reconciliation. Sources: rovai-ai:w-usage.
Compute cost [unsupported] No compute fees. Sources: rovai-ai:readme.
Refresh delay [supported] Usage flushes at most every 4 seconds; late numbers land on a periodic task; Refresh updates the view. Sources: rovai-ai:monitoring, rovai-ai:w-usage.
Alerts [unsupported] No budget alerts. Sources: rovai-ai:w-usage.
No compute-aware scheduling; per-teammate FIFO queues and one claimed prefix per run; Missions share a worktree; no CPU/RAM limits or check reuse.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
Not separately cross-checked
concurrency: FIFO per teammate; no global limit documented.
cpu_ram: No CPU/RAM limits.
backpressure: Work waits behind the current run; no resource-based backpressure.
host_awareness: No host resource awareness.
check_reuse: No checks run by the product.
check_deduplication: No check deduplication.
bounded_tests: No test runner.
fairness: FIFO prefix claims give ordering within a teammate.
Single-owner instance: one login token; multiple browsers can open the same workspace but there are no users, presence, comments or attribution among humans; Feishu/DingTalk channels carry requests from chat. Teammates (agents) share conversations.
Scope
Surfaces: Browser workspace (Desktop Web access or standalone Server), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
visibility: Any browser with the token sees the same instance; no per-user views.
discussion: Humans and agents share a conversation; only one human owner exists.
comments: No human comments system.
presence: No presence.
coprompting: Single owner; no multi-human prompting.
ordering: FIFO delivery ordering for messages.
conflicts: No multi-human conflict handling.
attribution: Messages carry author identity (owner or teammate).
11 atomic answers
Visibility [partial] Any browser with the token sees the same instance; no per-user views. Sources: rovai-ai:w-server-maint.
Discussion [partial] Humans and agents share a conversation; only one human owner exists. Sources: rovai-ai:w-collab.
Comments [unsupported] No human comments system. Sources: rovai-ai:srch-team.
Permissions [unsupported] No sharing permissions; "Do not share the Token as if it were a limited project invitation". Sources: rovai-ai:w-server-maint.
Plan restrictions [supported] Free; no plans. Sources: rovai-ai:readme.
Owner online [partial] The host must stay running; the owner's browser need not. Sources: rovai-ai:w-server-maint.
Single-owner instance with a login token; HTTPS or Tailscale protects the route, not separate roles; no SSO, membership or roles; channel accounts handled by provider apps.
Scope
Surfaces: Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
organization_roles: No organizations or roles; single-owner instance.
project_roles: No project roles.
task_roles: Task lead and assignee permissions among teammates; humans are one owner.
membership: No membership.
sso: No SSO.
provisioning: No provisioning.
unauthorized_access: Token login and browser session; a Desktop listener accepts connections on host interfaces; no documented rotate-token command.
revocation: No documented token rotation or session revocation command.
9 atomic answers
Organization roles [unsupported] No organizations or roles; single-owner instance. Sources: rovai-ai:w-server-maint.
Project roles [unsupported] No project roles. Sources: rovai-ai:srch-team.
Task roles [partial] Task lead and assignee permissions among teammates; humans are one owner. Sources: rovai-ai:w-tasks.
Approvals come from the Agent capability (native choices, one request at a time) and are surfaced in a pending area with exact command and scope; allow once or scope per the Agent; filesystem and approval policy are set per teammate; highest permission mode exists. Not product-enforced policy.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
tool: Tool and MCP approvals depend on the Agent.
network: Network actions appear as approvals when the Agent asks (e.g. localhost server).
secret: No secret approvals of Rovai's own.
delivery: Merging code is separate and manual; no delivery approval.
approver_identity: The single owner; no approver record beyond the instance.
lifetime: Scope is the Agent's native option (once or broader).
revoke: Stop the run; standing approvals belong to the Agent.
tunnels: Tailscale/proxy are the user's own network arrangements.
14 atomic answers
Command [supported] Approval cards show the exact command, working directory and native choices. Sources: rovai-ai:w-approvals.
SQLite-backed history of messages, deliveries, runs, tasks, missions, memory versions and approvals with evidence provenance; single owner; no export or integrity model beyond local storage.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
approvals: Approval requests and choices appear in the run record.
decisions: Memory agreements and Mission explanations.
cross_user_persistence: Single owner.
export: JSON export for usage; no general export.
mutability: Local SQLite database; evidence is authoritative but editable by the host user.
Files Changed cards per run and Mission cumulative changes with per-file diffs, file preview beside the conversation, Run overview, reviewer teammates and follow-up requests; no review queue or comment threads on code.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Browser workspace (Desktop Web access or standalone Server) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
pending_reviews: Pending approvals indicator; no review queue.
comments: No inline code comments.
request_changes: Send a bounded follow-up to the implementer; no formal object.
shipping_guidance: No shipping guidance; merging is separate and manual.
effort: No reviewer effort measurements.
missed_evidence: Files Changed appears only when reliable change evidence exists; unavailable diff is flagged as a missing view.
keyboard: Primary desktop surfaces support keyboard operation and focus (PRODUCT.md).
narrow_screen: Responsive browser UI and minimum 1040x700 desktop window.
Change understanding [supported] Expand changed files to read diffs and compare with the teammate's summary. Sources: rovai-ai:w-files.
Evidence navigation [supported] Open file links, steps and Files Changed from the run. Sources: rovai-ai:w-files, rovai-ai:w-exec.
Comments [unsupported] No inline code comments. Sources: rovai-ai:w-files.
Request changes [partial] Send a bounded follow-up to the implementer; no formal object. Sources: rovai-ai:w-collab.
Shipping guidance [unsupported] No shipping guidance; merging is separate and manual. Sources: rovai-ai:readme.
Effort [unknown] No reviewer effort measurements. Sources: rovai-ai:readme.
Missed evidence [partial] Files Changed appears only when reliable change evidence exists; unavailable diff is flagged as a missing view. Sources: rovai-ai:w-files.
Keyboard [partial] Primary desktop surfaces support keyboard operation and focus (PRODUCT.md). Sources: rovai-ai:product.
Context switching [supported] Preview opens beside the conversation; panel can pin right, float or sit at the bottom. Sources: rovai-ai:w-exec, rovai-ai:w-files.
Single-owner instance: another browser with the token continues the same conversation, runs and files; there is no second user or ownership transfer; separate instances do not migrate.
Scope
Surfaces: Browser workspace (Desktop Web access or standalone Server), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
ongoing_takeover: Another browser continues the same conversation with the owner's token.
permissions: No separate permissions.
ownership: No ownership transfer.
attribution: Single owner identity.
context_loss: Separate instances do not sync; moving to another host loses Desktop data.
harness_transfer: Change a teammate's Agent for future runs; no transcript transfer.
12 atomic answers
Ongoing takeover [partial] Another browser continues the same conversation with the owner's token. Sources: rovai-ai:w-desktop-web.
Same task [supported] Same conversation, Run and files on the host. Sources: rovai-ai:w-desktop-web.
Session history [supported] Existing workspace through a browser. Sources: rovai-ai:w-remote.
Decisions [supported] Memory and Mission records are in the instance. Sources: rovai-ai:w-memory.
Artifacts [supported] Files remain on the host. Sources: rovai-ai:w-remote.
Retained outputs: file previews (HTML, Markdown), Files Changed diffs per run and Mission, execution history and steps, message attachments; no screenshots, recordings or artifact gallery.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
screenshots: Message attachments and runtime images; no screenshot capture.
recordings: No recordings.
previews: File preview and system/browser action for rendered HTML.
test_outputs: Command steps and results in the Run panel.
gallery: No cross-task artifact gallery.
revision_links: Cumulative changes compare a recorded base with the current worktree; refresh is not a historical snapshot.
prompt_leakage: Usage monitoring excludes prompts, completions and tool output; model requests follow the Agent's provider.
Revision links [partial] Cumulative changes compare a recorded base with the current worktree; refresh is not a historical snapshot. Sources: rovai-ai:w-missions.
Persistence [supported] Persisted in the instance and project files. Sources: rovai-ai:product.
No product-run test, lint or CI gates; an independent reviewer teammate can be asked to review and run existing checks; success of a run does not certify content; no spec gates.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
Not separately cross-checked
tests: Teammates may run tests inside runs; no product test integration.
lint: No lint integration.
types: No type-check integration.
ci: No CI integration.
spec_checks: No spec checks.
required_checks: None enforced.
failure_blocking: Nothing blocks delivery; run completion does not mark Missions complete.
requirement_conflicts: No requirement store.
11 atomic answers
Tests [partial] Teammates may run tests inside runs; no product test integration. Sources: rovai-ai:w-collab.
Lint [unsupported] No lint integration. Sources: rovai-ai:srch-pr.
Types [unsupported] No type-check integration. Sources: rovai-ai:srch-pr.
Independent review [supported] Reviewer teammate with a read-only configuration is the documented pattern and review-duo skill. Sources: rovai-ai:w-collab, rovai-ai:cli-skill.
Git-aware only: observes Git state, manages Mission worktrees (prepare, verify, clean up) and shows cumulative changes; commits, pushes and PRs are done by the Agent; no PR integration.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
branches: Missions record a base; branch operations are the Agent's.
commits: Agents commit via their own tools; cumulative changes include committed work; no product commit action.
pull_requests: No PR creation or management.
comments: No PR comments.
ci_response: No CI response.
rebase: No rebase.
conflicts: No conflict workflow.
product_state: Mission workspace and Git observation are product state; no commit/PR state.
9 atomic answers
Branches [partial] Missions record a base; branch operations are the Agent's. Sources: rovai-ai:w-missions.
Commits [partial] Agents commit via their own tools; cumulative changes include committed work; no product commit action. Sources: rovai-ai:w-missions, rovai-ai:git-rs.
Pull requests [unsupported] No PR creation or management. Sources: rovai-ai:srch-pr.
Comments [unsupported] No PR comments. Sources: rovai-ai:srch-pr.
Ci response [unsupported] No CI response. Sources: rovai-ai:srch-pr.
The product has no approve, merge, release or deploy authority; merging code is a separate manual action from finishing a run or completing a Mission.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
approve: No delivery approval.
merge: No merge code.
release: No release tooling.
deploy: No deployment tooling.
exact_revision: No revision-bound approval.
authority_enforcement: No authority enforcement.
6 atomic answers
Approve [unsupported] No delivery approval. Sources: rovai-ai:readme.
Missions track cumulative changes against a recorded base and explicit status; run completion, Mission completion and merging are separate; no commit/PR/build/deployment links or rollback feature (stop is not undo).
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
commit_links: Cumulative changes include committed work against the recorded base.
pr_links: No PR links.
build_links: No build links.
deployment_links: No deployment links.
delivery_status: Mission business status is manual and distinct from execution.
rollback: Stop is not undo; changes to the Mission worktree remain until cleaned up.
revert: No revert feature.
stale_approval_invalidation: No approvals of delivery.
9 atomic answers
Commit links [partial] Cumulative changes include committed work against the recorded base. Sources: rovai-ai:w-missions.
No batch review or shipping queue; the Mission board and Run overview give per-Mission status, and cleanup is an explicit safety-checked action.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
review_queue: No review queue.
summaries: Teammate replies and run cards summarize work.
grouping: Board groups Missions by status.
risk_size_filters: No risk or size filters.
findings: Reviewer replies are retained.
artifacts: Files Changed per run.
request_changes: Follow-up request to the implementer.
MIT for the whole monorepo (Rust host and core, Electron Desktop, web UI, website) at v0.5.12; no closed or hosted components were found. Release binaries are built from the repository.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Fully self-hosted: Desktop or Server on user hardware with local SQLite data and no vendor account; separate instances keep separate data and do not migrate; data export is limited to usage JSON.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
vendor_dependencies: None for Rovai; Feishu/DingTalk channels use their providers; update checks use GitHub releases.
export: JSON export of usage entries; no export of conversations documented.
import: No import; opening a new Server does not bring Desktop data.
formats: SQLite data directory and project files.
history_portability: Remote access neither migrates nor synchronizes data between instances.
8 atomic answers
Self hosting [supported] Rovai Server and Desktop Web run on user-controlled hosts. Sources: rovai-ai:w-server.
Vendor dependencies [partial] None for Rovai; Feishu/DingTalk channels use their providers; update checks use GitHub releases. Sources: rovai-ai:w-channels, rovai-ai:updates.
Export [partial] JSON export of usage entries; no export of conversations documented. Sources: rovai-ai:w-usage.
Import [unsupported] No import; opening a new Server does not bring Desktop data. Sources: rovai-ai:w-remote.
Projects, credentials and Agent configuration stay on the host; model requests follow each Agent's provider ("a local workspace is not a guarantee that all inputs stay on the computer"); no telemetry or analytics code was found; Desktop checks GitHub for updates; usage metrics exclude prompts, completions and credentials; Memory has an open opt-out request.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
prompts: Prompts go to the Agent and its provider; Rovai stores messages locally.
retention: Derived usage rows are deleted after 45 days; conversation facts persist.
deletion: Conversation deletion exists as an architecture topic; account deletion is not applicable.
training: Provider training policies belong to each Agent vendor.
subprocessors: None for Rovai; channel providers and model providers are chosen by the user.
telemetry_disclosure: No telemetry statement found beyond the privacy-minimized usage reader; update service documented.
telemetry_controls: Controls for update checks are not documented in the pages read; no telemetry exists to disable.
15 atomic answers
Code data [supported] Files stay on the execution host; Agents send code to their configured model service. Sources: rovai-ai:product.
Prompts [partial] Prompts go to the Agent and its provider; Rovai stores messages locally. Sources: rovai-ai:product.
Credentials [supported] Workspaces, Runtime configuration and credentials remain user-owned and local; secrets are kept out of evidence. Sources: rovai-ai:product, rovai-ai:security.
Logs [supported] Local diagnostics and run records. Sources: rovai-ai:product.
Artifacts [supported] Local project files. Sources: rovai-ai:product.
Retention [partial] Derived usage rows are deleted after 45 days; conversation facts persist. Sources: rovai-ai:monitoring.
Deletion [partial] Conversation deletion exists as an architecture topic; account deletion is not applicable. Sources: rovai-ai:product.
Training [unknown] Provider training policies belong to each Agent vendor. Sources: rovai-ai:w-agent-setup.
Residency [supported] Wherever the host runs; provider residency is the Agent's. Sources: rovai-ai:w-remote.
Subprocessors [partial] None for Rovai; channel providers and model providers are chosen by the user. Sources: rovai-ai:w-channels.
Telemetry disclosure [partial] No telemetry statement found beyond the privacy-minimized usage reader; update service documented. Sources: rovai-ai:updates, rovai-ai:srch-tele.
Telemetry controls [unknown] Controls for update checks are not documented in the pages read; no telemetry exists to disable. Sources: rovai-ai:updates.
Free and open source (MIT) with no paid plan or hosted service; Agent subscriptions and model usage are separate.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
credits: No credits.
trial: No trial needed.
overages: None.
12 atomic answers
Currency [supported] Not applicable: no price. Sources: rovai-ai:readme.
Interval [supported] Not applicable. Sources: rovai-ai:readme.
Rovai costs $0 for one, two or five developers; Agent and model costs and optional host costs are separate and unmodeled. It is single-owner, so five people would run five instances.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
two_people: $0, but no two-person collaboration within one instance.
five_people: $0; five separate instances.
task_assumptions: No fixed workload published; the catalog invents none.
tokens: Token spend is provider-specific.
compute_hours: User's own host.
variable_cost: Agent usage varies.
unknown_cost: Agent provider and host costs unknown to Rovai.
Frequent releases (v0.5.9 to v0.5.12 in a week) with bilingual notes, private vulnerability reporting, GitHub issues and a WeChat community; no response commitments or status page; Linux Agent support and some platforms are preview.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
release_stage: 0.5.x with preview platforms for some Agents and Linux.
support: GitHub issues, private security reports and a WeChat group.
Everything is free (MIT) with no trial, card or caps; BYO Agent subscriptions, BYO remote host (Server), phone browser access, Missions and review work for free; two-person collaboration does not exist.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64), Rovai Server (rovai-server host package for macOS, Linux and Windows) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
trial: No trial needed.
credits_trial: No credits.
paid_only: No paid-only tier.
early_access: Some Agent adapters and Linux are preview.
Fully transparent: free, no plans, no markups; third-party Agent costs are separate.
Scope
Surfaces: Rovai Desktop (Electron app, macOS arm64/x64 and Windows x64) · Version: v0.5.12 (commit 120af14, released 2026-11-12) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
9 atomic answers
Units [supported] No billing units. Sources: rovai-ai:readme.
Included limits [supported] No limits. Sources: rovai-ai:readme.
Desktop and web workspace for lasting teams of coding agents with shared conversations, missions, scheduled work, memory, MCP, remote browser access and Feishu/DingTalk channels; MIT.
Kind
documented
Locator
capability table, Get started, How it fits together
Teammates use the built-in `rovai` CLI to send ThreadMessages (including parallel requests with repeated --to), read members, threads, execution state, Missions and memory, and update their own profile.
Desktop packages for macOS arm64/x64 and Windows x64; Server packages for macOS arm64/x64, Windows x64 and Linux x64 GNU; Linux agent support keeps a preview boundary.
Toggle in Settings → Remote connection; token login; listener accepts host network interfaces; QR login option; web service must be re-enabled after app restart.
Published native packages for macOS arm64/x64, Linux x64 glibc 2.35+, Windows x64; installer script with SHA256SUMS; ordinary user account; default data directory ~/.rovai-server; Channels not provided.
Notes add conversation fork and Kilo agent; assets are Desktop dmg/zip for macOS arm64/x64, Windows x64 exe, and Server packages for Linux, macOS and Windows with install scripts and SHA256SUMS.
Desktop workspace for long-lived agent teams that drives locally installed runtimes; SQLite core facts; usage is sparse and source-qualified; workspaces and credentials stay user-owned.
Kind
documented
Locator
Platform, Product Purpose, Capabilities and Constraints
Mission links a goal to one team conversation and workspace; Git projects get a managed shared worktree from current local HEAD; cumulative changes; start is separate from business status.
Public ThreadMessage creates waiting Deliveries per explicit target; a scheduler claims a FIFO prefix into an immutable AgentRun with a frozen context manifest.
Mentions are structured recipients; implementer and reviewer example; agent-to-agent handoff is a separate supported path; avoid two writers on one file.
Tasks record owner, scope and status (Pending, In progress, Blocked, Complete, Cancelled); creating a task does not start an Agent; cancel does not stop a run.
Host-based recurring requests: each occurrence creates its own conversation; overlapping occurrences are skipped; approvals cancel unattended runs; missed runs do not replay; not a cloud service.
Shared, teammate and between-teammate memory with scopes, retrieval keys, revisions and candidate review; agents may write their own teammate memory directly.
Reviewer bounded search at pinned commit: 2 file hit(s), reviewed: crates/rovai-core/src/application/config.rs, crates/rovai-core/src/application/transport.rs. Both hits are the word "embedding" in comments about the embedding process or Host, not a code index.
Reusable resident agent-runtime processes are capped at 20 per teammate bucket and 200 globally, with a 30-minute idle TTL and a 60-second sweep; when resident capacity is full a run starts as a non-resident Burst process, so the caps bound process reuse rather than concurrent runs.
RUN_FACTS sets new_a2a_dispatch_allowed and new_a2a_target_contact_allowed to false once a2a_depth is 5 or more or the a2a run count is 16 or more, so agent-to-agent dispatch is bounded per conversation chain.
Reviewer bounded search at pinned commit: 7 file hit(s), reviewed: crates/rovai-core/src/application.rs, crates/rovai-core/src/member_avatar.rs, crates/rovai-core/src/diagnostics.rs, apps/desktop/src/main/dingtalk-developer-gateway.ts, apps/desktop/src/renderer/src/ThreadDetailPopover.tsx, apps/desktop/src/renderer/src/DiagnosticsCenter.tsx, packages/contracts/src/index.ts. Every hit is a substring match inside the identifiers DiagnosticGroup, dynamicGroup (cGroup) and avatarLimit (rlimit) or a GIF decoder memory limit in member_avatar.rs; none reads host CPU, memory or load.
File preview beside the conversation, per-run Files Changed cards and Mission cumulative changes; an unavailable diff is a missing view, not proof of no changes.