51 of 51 criteria have a sourced finding; 42 were also cross-checked by a second agent. Every comparison that includes AI Maestro 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.
Browser dashboard (desktop, tablet and phone layouts, installable as a PWA) served by the host, a launcher command, and a separate iPhone app; macOS and Linux hosts, Windows via WSL2; no desktop-native app or editor extension; Android app is an email-invited beta.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents), AI Maestro iPhone app and Android beta (closed source) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
Windows only through WSL2.
Mobile apps are closed source and Android is an invite-only beta.
No editor extension.
9 atomic answers
Desktop os [partial] The server runs on macOS 12+ and Linux natively and on Windows only inside WSL2; it ships no native desktop app, only a Dock launcher that opens the dashboard. Sources: ai-maestro:readme, ai-maestro:reqs, ai-maestro:quickstart.
Mobile apps [partial] An iPhone app is on the App Store (iOS 15.1+, 30 days free then $19.99 one-time per the vendor page); Android 7+ is an invite-only beta joined by emailing support. Neither app is in the public repositories. Sources: ai-maestro:iphonepage, ai-maestro:appstore, ai-maestro:srch-clients.
Browser [supported] Full dashboard at http://localhost:23000 with chat, terminal, board, matrix, attention and team pages; installable as a PWA over https or localhost. Sources: ai-maestro:readme, ai-maestro:manifest.
Mobile browser [supported] Responsive phone layouts (MobileDashboard, MobileChatView) and Add to Home Screen; the README shows phone Matrix, home and board screens. Sources: ai-maestro:readme, ai-maestro:product, ai-maestro:manifest.
Cli [supported] The installer adds an ai-maestro command (open, start, stop, status) and the plugin ships aimaestro-agent.sh, aimaestro-task.sh and AMP scripts. Sources: ai-maestro:readme, ai-maestro:plugscripts.
Editor extensions [unsupported] Bounded search finds no VS Code or JetBrains extension; the Claude Code mods are plugins for that CLI, not editor extensions. Sources: ai-maestro:srch-clients.
Task actions [partial] Create agents, send chat, steer through the terminal, answer approval and question cards, move cards and schedule work from the dashboard; review of code changes and resume of an old session beyond waking the agent are not separate flows. Sources: ai-maestro:readme, ai-maestro:programs, ai-maestro:tasksdoc.
Notifications [partial] Push to the iPhone app through Expo (thin payload, 60 s cooldown); in-dashboard status, unread counts and Attention page; browser-level notifications are not documented. Sources: ai-maestro:pushexpo, ai-maestro:readme.
Cross device continuity [supported] Agents live in tmux on the host and the dashboard is stateless in the browser, so any device on the network or Tailscale sees the same agents; mesh peers show all hosts. Sources: ai-maestro:concepts, ai-maestro:security.
Runs coding CLIs inside tmux: Claude Code, Codex CLI and Grok Build with chat, status and approval cards; Pi, Copilot CLI and OpenClaw with reduced support; any other terminal agent in the terminal tab. Also an embedded Claude Agent SDK chat path.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
Support depth differs per CLI.
No custom adapter interface; each reader is built into the server.
5 atomic answers
Harnesses [supported] Claude Code, Codex CLI, Grok Build (first-class); Pi, GitHub Copilot CLI, OpenClaw recognised; Cursor, Gemini CLI, OpenCode and others only as terminal programs; Aider deprecated. Sources: ai-maestro:programs, ai-maestro:prognames.
Model providers [partial] Model choice is left to each CLI; the product only passes a model setting (--model) for some and documents Claude Code on Vertex AI and Pi with Gemini on Vertex. Sources: ai-maestro:programs, ai-maestro:programlaunch.
Custom adapters [partial] Any terminal program can be set as an agent program and typed into; transcript readers (transcript-codex/grok/pi/copilot.mjs) are built-in modules rather than a plugin API. Sources: ai-maestro:programs, ai-maestro:prognames.
Session integration [supported] Default path controls the CLI through tmux (paste-buffer and key presses) plus hooks and transcript reading; the Streaming tab embeds the Claude Agent SDK with canUseTool approvals. Sources: ai-maestro:chatarch, ai-maestro:pkg, ai-maestro:hook.
Mixed harness operation [supported] One dashboard and mesh can host Claude Code, Codex and Grok Build agents side by side that message each other over AMP. Sources: ai-maestro:programs, ai-maestro:readme.
Uses each CLI's own login or API key; AI Maestro holds no model credentials. Subscription use follows whatever the CLI supports; no vendor credits.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
No account switching in AI Maestro itself; each host uses the login its CLI holds.
7 atomic answers
Agent subscriptions [supported] Agents run the user's logged-in CLI (Grok needs SuperGrok/X Premium+, Copilot needs a Copilot subscription, Claude Code its own login); the README says to bring your own agent subscriptions. Sources: ai-maestro:readme, ai-maestro:programs.
Api accounts [supported] CLIs started with their own API keys work; Claude Code on Vertex AI and Pi on Vertex are documented with helper scripts. Sources: ai-maestro:programs.
Vendor credits [unsupported] No vendor credits, token sales or hosted tier exist ("no paid tier", "Is there a hosted version? Not yet"). Sources: ai-maestro:readme, ai-maestro:srch-billing.
Local models [partial] Pi can use a local OpenAI-compatible server and the repository tests used a fake local model, but local models are a property of the CLI, not an AI Maestro feature. Sources: ai-maestro:programs.
Account switching [unsupported] Bounded search of lib, services, docs and README finds no multi-account switching; each host uses whichever login its CLI holds. Sources: ai-maestro:srch-acct.
Vendor account requirement [supported] No AI Maestro account or login is required ("No account, no telemetry"). Sources: ai-maestro:readme.
Provider restrictions [partial] Documents per-CLI login and subscription needs (Grok, Copilot) but no restrictions imposed by AI Maestro on third-party subscription use. Sources: ai-maestro:programs.
One-line installer (curl | bash) for macOS, Linux or WSL2 needing Node 24, tmux and about 4 GB RAM to build; first agent via wizard, CLI or by starting a tmux session; local directory, local git and GitHub repos all work; BYO VPS and AWS paths documented.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
Manual steps for installing each CLI and logging in on every host.
7 atomic answers
First task setup [supported] Install, open localhost:23000, create an agent with the guided wizard, wake it and chat; or run aimaestro-agent.sh create; any existing tmux session is auto-discovered. Sources: ai-maestro:readme, ai-maestro:quickstart.
No git [supported] An agent is a working directory plus tmux session; git is not required to create or run one. Sources: ai-maestro:concepts.
Local git without remote [supported] Repos are discovered by .git presence and read with git config; a remote origin is optional. Sources: ai-maestro:concepts.
Hosted git [partial] A GitHub token can be passed to Docker and EC2 agents for push; the host environment otherwise uses the user's existing git credentials. Sources: ai-maestro:docker, ai-maestro:tfreadme.
Prerequisite guidance [supported] README lists Node 24, tmux, RAM, WSL2 steps and Linux sizing; verify-installation.sh and the Doctor check readiness. Sources: ai-maestro:readme.
Managed cloud setup [unsupported] No vendor-managed cloud exists ("Is there a hosted / cloud version? Not yet"). Sources: ai-maestro:readme.
Byo vps onboarding [supported] Documented Linux deployment guide, headless worker mode, remote-install.sh and the Terraform EC2 and ECS Fargate modules. Sources: ai-maestro:readme, ai-maestro:tfreadme, ai-maestro:cloud.
Local REST API (no auth), WebSocket, shell scripts, signed webhooks and AMP API keys; no SDK; the product does not expose an MCP server for controlling the dashboard, only an MCP channel server for pushing messages into Claude Code.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
Unauthenticated dashboard API by design.
No SDK.
8 atomic answers
Cli [supported] aimaestro-agent.sh (create, rename, hibernate, wake, delete), aimaestro-task.sh, amp-* scripts and the ai-maestro launcher. Sources: ai-maestro:plugscripts, ai-maestro:readme.
Api [supported] Next.js REST routes under app/api for agents, chat, messages, tasks, teams, hosts, schedules, webhooks and a versioned AMP API at /api/v1. Sources: ai-maestro:pkg, ai-maestro:tasksdoc.
Sdk [unsupported] No client SDK is published in the inspected repositories. Sources: ai-maestro:srch-clients.
Mcp server [partial] amp-channel-server.mjs is an MCP server that feeds inbox events into a running Claude Code session; no MCP server exposes agent control to other tools. Sources: ai-maestro:channel.
Webhooks [supported] Webhook subscriptions with whsec_ signing secrets for agent and email events. Sources: ai-maestro:webhooks.
Operations [supported] Create, inspect, chat/steer, hibernate, wake, delete, message and task operations exist in the API; stop is hibernate or tmux kill; code review is not an API operation. Sources: ai-maestro:tasksdoc, ai-maestro:programs.
Authentication [partial] The dashboard API has no authentication by design; AMP /api/v1 uses hashed API keys with rotation and cross-host AID/Ed25519 signatures. Sources: ai-maestro:security, ai-maestro:ampauth.
Stability [partial] The dashboard API has no stability or deprecation policy in README, docs or SECURITY.md (bounded grep); the product is pre-1.0 (0.106.x) and only the AMP routes are namespaced /api/v1. Sources: ai-maestro:srch-stab, ai-maestro:pkg.
Agents are launched CLIs, so MCP and plugin support is inherited from them; AI Maestro adds a Claude Code plugin with 14 skills (Agent Skills standard), a plugin builder and a marketplace page, and an MCP channel server.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
Mostly inherited from each CLI.
7 atomic answers
Mcp client [partial] MCP client features belong to the underlying CLI; AI Maestro only wires its own channel MCP server into Claude Code and offers Pi tool profiles that exclude mcp__* tools. Sources: ai-maestro:channel, ai-maestro:programs.
Plugins [supported] Claude Code plugin (marketplace ai-maestro-plugins, 14 skills), plugin builder with 150+ personalities, and Claude Code mods for inbox and secrets. Sources: ai-maestro:readme, ai-maestro:plugskills.
Custom tools [partial] Skills and scripts add tools for agents; Pi gets a copied extension (amp_*); no generic custom-tool registry for all programs. Sources: ai-maestro:plugskills, ai-maestro:programs.
Instructions import [partial] Team-level markdown instructions and per-agent files; CLAUDE.md and AGENTS.md handling is each CLI's; the installer writes a managed block into Pi AGENTS.md. Sources: ai-maestro:teamtype, ai-maestro:programs.
Authentication [partial] Integration credentials go through aim-secret (vault) or each tool's own login. Sources: ai-maestro:secrets.
Permission scope [partial] Per-program permission modes and a Pi messaging tool profile; skills have no separate scope model. Sources: ai-maestro:progchoices, ai-maestro:programs.
Integration owner [supported] Skills and scripts come from the AI Maestro plugin; MCP/tool wiring for other CLIs is owned by those CLIs; gateways are a separate repository. Sources: ai-maestro:plugskills, ai-maestro:gateways.
Web app: Next.js/React with xterm.js terminal and a Node server (tmux via node-pty) plus a PWA manifest; no Electron or Tauri client; the iPhone and Android apps are closed and their stack is not published (vendor page calls the chat "native").
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents), AI Maestro iPhone app and Android beta (closed source) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 client implementation not public.
4 atomic answers
Desktop implementation [supported] No desktop client: the dashboard is a Next.js app with xterm.js/wterm terminals served by server.mjs; a Mac launcher only opens the browser. Sources: ai-maestro:pkg, ai-maestro:srch-clients.
Mobile implementation [unknown] The iPhone and Android apps are not in the public repositories; the App Store listing and the vendor page do not state the framework (push uses Expo, which suggests but does not prove React Native). Sources: ai-maestro:iphonepage, ai-maestro:appstore, ai-maestro:pushexpo.
Hybrids [supported] Web UI plus PWA install; no hybrid wrapper in the repository. Sources: ai-maestro:manifest.
Implementation evidence [partial] Pinned package.json and source cover the web app; mobile builds are closed. Sources: ai-maestro:pkg, ai-maestro:appstore.
Local card creation on agent, team and project boards, prompt intake through chat, and import of an existing task list through a CLI; no GitHub/Linear/Jira issue import and no chat or Git triggers.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: aimaestro-task.sh import accepts a JSON task list; no tracker connector found.
chat git triggers: Slack/Discord/Email/WhatsApp gateways forward messages to agents; no Git or CI trigger found.
sync: No external tracker sync exists in the inspected code.
external roundtrips: Not applicable: nothing is synced to an external tracker.
9 atomic answers
Local creation [supported] Cards are created from the dashboard, aimaestro-task.sh add/park and the REST API. Sources: ai-maestro:tasksdoc.
Prompt intake [supported] Prompts go through chat or typed instructions; schedules can issue prompts. Sources: ai-maestro:chatarch, ai-maestro:sched.
External import [partial] aimaestro-task.sh import accepts a JSON task list; no tracker connector found. Sources: ai-maestro:tasksdoc, ai-maestro:srch-pr.
Chat git triggers [partial] Slack/Discord/Email/WhatsApp gateways forward messages to agents; no Git or CI trigger found. Sources: ai-maestro:gateways, ai-maestro:srch-pr.
Authoritative tracker [supported] The JSON board files in the user's ~/.aimaestro are authoritative; no external tracker is linked. Sources: ai-maestro:taskreg.
Stable ids [supported] Cards have UUIDs (short prefix shown in instructions). Sources: ai-maestro:taskreg.
Sync [unsupported] No external tracker sync exists in the inspected code. Sources: ai-maestro:srch-pr.
Cards carry description and a result field; a planning skill keeps persistent plan, findings and progress files; no per-task acceptance criteria or revision history.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: The planning skill writes plan/findings/progress files for long tasks, and Claude plan mode is a permission mode; cards hold a description only.
acceptance criteria: No acceptance-criteria field on cards in the inspected types.
revisions: Cards store comments but no edit history.
implementation links: Card origin links the agent, session, repo and thread that created it; commits trace back to sessions.
7 atomic answers
Plans [partial] The planning skill writes plan/findings/progress files for long tasks, and Claude plan mode is a permission mode; cards hold a description only. Sources: ai-maestro:plugskills, ai-maestro:progchoices.
Acceptance criteria [unsupported] No acceptance-criteria field on cards in the inspected types. Sources: ai-maestro:tasktype.
Editors [supported] Card detail and board edit description and fields in the dashboard. Sources: ai-maestro:tasksdoc.
Revisions [unsupported] Cards store comments but no edit history. Sources: ai-maestro:tasktype.
Implementation links [partial] Card origin links the agent, session, repo and thread that created it; commits trace back to sessions. Sources: ai-maestro:tasksdoc, ai-maestro:commitrace.
Amendment [supported] A card can be edited while its agent works; instructions can be resent. Sources: ai-maestro:tasksvc.
Restart persistence [supported] Boards are JSON files on disk that survive server restarts. Sources: ai-maestro:taskreg.
Flat cards with status, blockedBy dependencies (cycle-checked, readiness derived), kinds and quadrants; teams and projects group cards; no epics or arbitrary parent/child hierarchy and no dependency-gated dispatch.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
epics: No epic or parent field on cards; teams and projects are containers.
hierarchy: Card hierarchy is not modelled; agents have a manager hierarchy and org chart instead.
readiness: isBlocked is computed from incomplete blockers and completing a card reports newly unblocked cards; there is no ready queue.
enforcement: Blockers are shown, not enforced: no code stops an agent from starting a blocked card.
Hierarchy [unsupported] Card hierarchy is not modelled; agents have a manager hierarchy and org chart instead. Sources: ai-maestro:tasktype, ai-maestro:agenttype.
Blockers [supported] blockedBy lists of card IDs with a derived blocks list and isBlocked flag. Sources: ai-maestro:taskreg.
Readiness [partial] isBlocked is computed from incomplete blockers and completing a card reports newly unblocked cards; there is no ready queue. Sources: ai-maestro:taskreg.
Enforcement [unsupported] Blockers are shown, not enforced: no code stops an agent from starting a blocked card. Sources: ai-maestro:taskreg, ai-maestro:tasksvc.
State propagation [partial] Completion returns the unblocked cards but does not move them. Sources: ai-maestro:taskreg.
Task to epic [unsupported] No epic concept to grow into. Sources: ai-maestro:tasktype.
Identity retention [not-applicable] Not applicable: cards cannot grow into epics; a card keeps its UUID, comments and history across edits. Sources: ai-maestro:taskreg, ai-maestro:tasktype.
Workflow constraints [supported] No rigid workflow is required; any status can be set. Sources: ai-maestro:tasktype.
Per-agent working directory with CLI instructions, team-level markdown instructions, 14 skills, a skill marketplace and per-agent opt-in long-term memory (memory cards and entity graph with recall) plus code graph and generated docs.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
Memory is opt-in per agent and team memory is not shared by default.
10 atomic answers
Instructions [supported] Team instructions field and each CLI's own instruction files; managed AGENTS.md block for Pi. Sources: ai-maestro:teamtype, ai-maestro:programs.
Skills [supported] 14 skills in the Agent Skills format installed with the plugin plus a marketplace service. Sources: ai-maestro:plugskills, ai-maestro:readme.
Project docs [supported] Docs skill and DocumentationPanel index generated documentation of the agent codebase. Sources: ai-maestro:readme.
Memory [supported] Nightly consolidation of conversations into memory cards and an entity graph, with recall injected on prompts, per agent opt-in. Sources: ai-maestro:ltm.
Scope [supported] Memory and code graph are indexed per agent and what it owns; team memory is not shared by default. Sources: ai-maestro:readme, ai-maestro:ltm.
Provenance [supported] Each memory statement is backed by the passages it came from. Sources: ai-maestro:readme, ai-maestro:ltm.
Correction [supported] A PATCH route edits a memory's content, category and context, and memories are rebuilt from sources by nightly consolidation. Sources: ai-maestro:memroute, ai-maestro:ltm.
Deletion [supported] A DELETE route removes one memory by id and MemoryViewer has a per-memory delete button. Sources: ai-maestro:memroute, ai-maestro:memviewer.
Freshness [supported] Ended relations are marked and recurrence weights knowledge; nightly consolidation keeps it current. Sources: ai-maestro:readme, ai-maestro:ltm.
Takeover continuity [supported] Memory and agent identity survive new sessions and agent transfer between hosts. Sources: ai-maestro:readme, ai-maestro:concepts.
Code graph (CozoDB, ts-morph, delta indexing), generated docs search, memory search and conversation search per agent; images through each CLI; no cross-repo knowledge connectors or review-time enforcement.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 knowledge: Plugin personalities and gateways; no connector to external knowledge bases in core.
multimodal inputs: Agent uploads and AMP attachments exist; image reasoning depends on the CLI.
citations: Memory cards cite passages; graph answers list callers and dependencies.
conflict detection: No conflict-detection logic found.
Indexing [supported] Delta indexing of the codebase graph and transcripts (index-delta route). Sources: ai-maestro:readme.
External knowledge [partial] Plugin personalities and gateways; no connector to external knowledge bases in core. Sources: ai-maestro:readme, ai-maestro:gateways.
Multimodal inputs [partial] Agent uploads and AMP attachments exist; image reasoning depends on the CLI. Sources: ai-maestro:srch-review.
Native cards with create, edit, comments, priority, quadrant, assignee, owners, due, kinds (idea/bug/research/task), Eisenhower Matrix, Attention page and Kanban; no labels, archive or search listed; ideas can be parked without starting work.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
label: kind and org labels exist; no free-form labels.
search: the Matrix filters by agent name only; no free-text card search was found.
archive: Cards are completed or deleted; no archive state.
project epic views: Agent, team and project boards and an Org view; no epics.
17 atomic answers
Create [supported] Create from dashboard, CLI and API. Sources: ai-maestro:tasksdoc.
Discuss [supported] Comments with operator or agent authors. Sources: ai-maestro:tasktype.
Prioritize [supported] priority and urgent/important quadrant. Sources: ai-maestro:tasktype.
Label [partial] kind and org labels exist; no free-form labels. Sources: ai-maestro:tasktype.
Assign [supported] Primary assignee plus up to 10 owners; project cards forward instructions to the assignee. Sources: ai-maestro:tasksdoc, ai-maestro:tasksvc.
Search [partial] The Matrix page filters cards by agent name or alias; no free-text search over card subjects or descriptions exists in the inspected components. Sources: ai-maestro:matrixpage.
Filter [supported] Matrix, Org and Attention views filter by quadrant, org, overdue and waiting. Sources: ai-maestro:tasksdoc.
Archive [unsupported] Cards are completed or deleted; no archive state. Sources: ai-maestro:tasktype.
Reopen [supported] Status can be moved back from completed. Sources: ai-maestro:tasktype.
Project epic views [partial] Agent, team and project boards and an Org view; no epics. Sources: ai-maestro:tasksdoc.
Idea parking [supported] `aimaestro-task.sh park` saves an idea card (kind idea) off the board. Sources: ai-maestro:tasksdoc.
Explicit dispatch [partial] Assignment forwards a typed instruction to the assignee on project boards; own-board cards need the agent to read them. Sources: ai-maestro:tasksvc.
Task session separation [partial] Cards are separate from sessions and carry origin ids, but a card does not own sessions. Sources: ai-maestro:tasksdoc.
Agents operate the same cards through aimaestro-task.sh and the REST API (add, list, due, owners, park, import, show, delete own board) with operator/agent attribution on comments; permissions are by board scope, not roles.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
claim: An agent can add a card to its own board or take owner status; no explicit claim lock.
children: No child cards.
attribution: Comments record operator or agent author; card field edits are not attributed.
permissions: Agents can only delete cards on their own board; no per-user permissions.
conflicts: JSON file read-modify-write with no lock or version check.
Conflicts [unsupported] task-registry reads and rewrites the whole JSON board synchronously with no lock, version check or merge, so the last writer wins. Sources: ai-maestro:srch-taskcas.
No living product specification: persistent docs, memory and planning files exist but without revisions, approvals or drift checks.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
persistent requirements: No requirements store separate from cards, memory and docs.
human edits: Not applicable beyond editing files.
agent edits: Planning skill files are agent-written but not a spec system.
revisions: No spec revisions.
9 atomic answers
Persistent requirements [unsupported] No requirements store separate from cards, memory and docs. Sources: ai-maestro:tasktype, ai-maestro:ltm.
Human edits [unsupported] Not applicable beyond editing files. Sources: ai-maestro:tasktype.
Agent edits [unsupported] Planning skill files are agent-written but not a spec system. Sources: ai-maestro:plugskills.
Any number of agents run concurrently, each in its own tmux session (or container/instance), across hosts; the sidebar and Zoom wall show working, needs-you, ready and offline states; no documented quota.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
quotas: No agent-count quota found; AMP rate limit is 60 messages per minute.
observed concurrency: Vendor reports 80+ agents and a Linux memory sizing doc; this catalog did not run it.
5 atomic answers
Concurrent sessions [supported] Each agent is a tmux session; 80+ agents are described across hosts. Sources: ai-maestro:readme, ai-maestro:concepts.
Concurrent tasks [supported] Each agent works on its own task concurrently. Sources: ai-maestro:readme.
Assignment of cards to agents and co-owners; agents own repositories by convention; no queue, routing engine or enforced exclusive ownership; diagnostics warn about shared folders.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
claims: Owners list marks shared responsibility; no exclusive claim.
queues: wake-queue and message queues exist for waking agents, not for work.
routing: Closed teams route external messages through the chief-of-staff.
duplicate prevention: Diagnostics warn when two agents share a folder; agents clone separately.
7 atomic answers
Human assignment [supported] Humans assign cards to agents and set a human manager in agent metadata. Sources: ai-maestro:tasksvc, ai-maestro:agenttype.
Agent assignment [supported] Agents add and assign cards via CLI/API. Sources: ai-maestro:tasksdoc.
Claims [partial] Owners list marks shared responsibility; no exclusive claim. Sources: ai-maestro:tasksdoc.
Queues [partial] wake-queue and message queues exist for waking agents, not for work. Sources: ai-maestro:tasksvc.
Routing [partial] Closed teams route external messages through the chief-of-staff. Sources: ai-maestro:teamtype.
Duplicate prevention [partial] Diagnostics warn when two agents share a folder; agents clone separately. Sources: ai-maestro:diag, ai-maestro:concepts.
Ownership enforcement [unsupported] Ownership is a convention; nothing prevents two agents from taking the same card. Sources: ai-maestro:taskreg.
Agents message each other with signed AMP messages (inbox, reply, wake), teams run split-pane meetings, agents have manager/chief-of-staff roles, and a skill lets an agent create and manage other agents; no parent/child spawn with returned results.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 harness effort: Model and program are set per agent at creation or wake; no per-delegation effort control.
results return: Replies come back as AMP messages; there is no structured result return.
live progress: Status and chat of any agent are visible; no parent-level child progress view.
12 atomic answers
Roles [supported] AgentRole manager, chief-of-staff, member, plus a manager link and org chart. Sources: ai-maestro:agenttype.
Model harness effort [partial] Model and program are set per agent at creation or wake; no per-delegation effort control. Sources: ai-maestro:progchoices.
Messages [supported] AMP signed messages with priority and types, delivered by Stop hook, channel or typing. Sources: ai-maestro:hook, ai-maestro:channel.
Results return [partial] Replies come back as AMP messages; there is no structured result return. Sources: ai-maestro:hook.
Cross harness [supported] Claude Code, Codex and Grok agents exchange AMP messages. Sources: ai-maestro:programs.
Orchestration owner [supported] Product-owned messaging and agent management skills; no hidden provider subagents. Sources: ai-maestro:plugskills.
Activation [supported] Installed with the plugin; messaging on by default. Sources: ai-maestro:readme.
Authentication [supported] AMP API keys and Ed25519 signatures; AID identity. Sources: ai-maestro:ampauth.
Live progress [partial] Status and chat of any agent are visible; no parent-level child progress view. Sources: ai-maestro:readme.
Retained results [supported] Messages persist in the inbox files and transcripts. Sources: ai-maestro:ampauth.
Followup [supported] Any agent can be messaged again after finishing. Sources: ai-maestro:readme.
Invocation [supported] Natural language ("send a message to backend") through the agent-messaging skill. Sources: ai-maestro:readme.
Per-agent cron schedules stored with the agent and fired by the dashboard server when the agent is idle; schedules travel with the agent; overlap skipped; no event triggers other than gateways and webhooks.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: Gateway messages and AMP messages wake agents; no repository event triggers.
completion conditions: No completion conditions on schedules.
retries: Error back-off instead of immediate retry; MAX 50 executions logged.
8 atomic answers
Schedules [supported] Cron expressions with timezone per schedule. Sources: ai-maestro:schedreg.
Events [partial] Gateway messages and AMP messages wake agents; no repository event triggers. Sources: ai-maestro:gateways, ai-maestro:hook.
Recurring tasks [supported] Morning triage, nightly checks and weekly reports are the documented use. Sources: ai-maestro:readme.
Completion conditions [unsupported] No completion conditions on schedules. Sources: ai-maestro:sched.
Scheduler location [supported] Runs inside the AI Maestro server on the host (60 s loop). Sources: ai-maestro:sched.
Chat and terminal steering, hibernate/wake, approval and question cards, Attention page and phone push; stop is hibernation or terminating the session; terminal and chat share the same tmux session.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: Hibernate stops the session; no mid-turn pause.
notifications: Push to the iPhone app and in-dashboard Needs you; Slack/Discord gateways.
escalation: Agent-to-human managers and the chief-of-staff role; no timed escalation.
visible errors: Delivery verification and Doctor checks; some failures only show in the terminal.
10 atomic answers
Pause [partial] Hibernate stops the session; no mid-turn pause. Sources: ai-maestro:readme.
Stop [supported] Hibernate or delete agent; terminal Ctrl+C. Sources: ai-maestro:readme.
Resume [supported] Wake resumes the agent (claude --continue style) with history. Sources: ai-maestro:readme.
Steer [supported] Chat or terminal at any time. Sources: ai-maestro:chatarch.
Pending decisions [supported] Questions and approvals appear as cards; Attention page lists them. Sources: ai-maestro:programs.
Escalation [partial] Agent-to-human managers and the chief-of-staff role; no timed escalation. Sources: ai-maestro:agenttype.
Visible errors [partial] Delivery verification and Doctor checks; some failures only show in the terminal. Sources: ai-maestro:chatarch.
Chat terminal parity [partial] The tmux chat path reads the terminal as text and is documented as inherently below 100% reliable. Sources: ai-maestro:chatarch.
Retained history [supported] Transcripts, search, export and playback tabs. Sources: ai-maestro:readme.
Dashboard, Zoom wall, Matrix, Attention, Org and Teams pages show agents and cards across hosts with status colours, due and overdue badges; no failed/reviewing lanes or epic dependency graph.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: Manager hierarchy and org view; no parent/child cards.
dependencies: blockedBy on cards; no dependency graph.
history: Playback and card timestamps; no fleet activity timeline.
workload: Matrix counts open cards per agent and host.
8 atomic answers
Fleet fixture [supported] Vendor screenshots use a demo company (Northwind Labs) with scripted agents; this catalog did not run a fixture. Sources: ai-maestro:readme.
No mandatory preset; natural-language messaging and teams; per-agent program, model and permission mode set at creation; war-room meetings show participating agents; no coordinator-run plan views.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: No epics.
presets: Wake dialog launch choices and plugin personalities act as presets; none required.
in run changes: Chat steering and card edits; no coordinator re-plan.
live roles: Team meeting tiles show agents; roles are manager/COS/member.
10 atomic answers
Quick task [supported] Chat with an agent or create a card. Sources: ai-maestro:chatarch.
Agents run in tmux on your host, on mesh peers you own (including headless workers), in Docker, or on EC2/ECS Fargate in your own AWS account; no vendor cloud. tmux sessions keep running when the browser closes; the host must stay on.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
managed cloud: No vendor-operated cloud ("Not yet").
resources: Docker agents default to 2 CPUs and 4g; Linux sizing guide for hosts.
expiry: EC2/ECS run until destroyed; Docker restart unless-stopped; no vendor lifetime.
12 atomic answers
Local [supported] tmux on the host running the dashboard. Sources: ai-maestro:readme.
Payer [supported] The user pays their cloud provider and CLI subscriptions. Sources: ai-maestro:readme.
Resources [partial] Docker agents default to 2 CPUs and 4g; Linux sizing guide for hosts. Sources: ai-maestro:docker, ai-maestro:readme.
Region [supported] The user picks the AWS region through the aws_region Terraform variable (default us-east-1); there is no vendor region because there is no vendor cloud. Sources: ai-maestro:tfvars.
Expiry [partial] EC2/ECS run until destroyed; Docker restart unless-stopped; no vendor lifetime. Sources: ai-maestro:docker.
Client disconnect [supported] tmux sessions persist after the dashboard or browser closes. Sources: ai-maestro:security.
Default tmux agents run as the host user in their own working directory (separate clone), not worktrees; Docker agents get a hardened container with CPU and memory limits; EC2 gives a dedicated instance. No documented network or secret boundary for tmux agents.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · Plan: Open source (MIT) · OS: Not specified · Execution modes: local, docker, cloud · 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: Agents integrate through git branches by convention; no product branch management.
worktree: The product deliberately does not use worktrees to isolate agents (README); agents may use them inside their clone.
vm: EC2 agents run on dedicated instances through Terraform; no local VM isolation.
os sandbox: Only launch flags of supported CLIs (e.g. Codex sandbox mode); tmux agents are unsandboxed.
Branch [partial] Agents integrate through git branches by convention; no product branch management. Sources: ai-maestro:readme.
Worktree [unsupported] The product deliberately does not use worktrees to isolate agents (README); agents may use them inside their clone. Sources: ai-maestro:readme, ai-maestro:diag.
Container [supported] Docker agent runtime with --cap-drop=ALL, no-new-privileges, noexec tmpfs, --cpus and --memory. Sources: ai-maestro:docker.
Vm [partial] EC2 agents run on dedicated instances through Terraform; no local VM isolation. Sources: ai-maestro:cloud, ai-maestro:tfreadme.
Network [unsupported] Docker publishes a port and adds host.docker.internal; no network policy. Sources: ai-maestro:docker.
Secrets [partial] aim-secret hides values from model context; file backend readable by the agent. Sources: ai-maestro:secrets.
Code separation [partial] Separate clones per agent; shared folders produce a warning. Sources: ai-maestro:diag.
Security boundary [partial] SECURITY.md and Pi docs state tools restrictions are not a security boundary. Sources: ai-maestro:security, ai-maestro:programs.
tmux persists sessions across dashboard restarts; wake resumes the CLI; transcripts reload for chat; agents can be transferred between hosts; context loss and manual repair for stuck panes are documented.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
worker crash: Hibernate/wake and Doctor checks; crashed CLIs need a wake.
expiry: Claude Code deletes transcripts after 30 days; AI Maestro rebuilds history from its message index for memory.
state reconstruction: Pending reconcile and chat-state files; Doctor.
context loss: Compaction and the 30-day transcript limit are cited.
12 atomic answers
Client disconnect [supported] tmux sessions persist after closing the dashboard. Sources: ai-maestro:security.
Worker crash [partial] Hibernate/wake and Doctor checks; crashed CLIs need a wake. Sources: ai-maestro:readme.
Service restart [supported] Agents are tmux sessions independent of the Node server and are rediscovered. Sources: ai-maestro:concepts.
Expiry [partial] Claude Code deletes transcripts after 30 days; AI Maestro rebuilds history from its message index for memory. Sources: ai-maestro:ltm.
Transcript reload [supported] Chat history reads each CLI transcript files. Sources: ai-maestro:programs.
Process reconnect [supported] Terminal re-attaches to tmux. Sources: ai-maestro:concepts.
Task resume [supported] Wake continues the agent session. Sources: ai-maestro:programs.
State reconstruction [partial] Pending reconcile and chat-state files; Doctor. Sources: ai-maestro:hook.
Context loss [partial] Compaction and the 30-day transcript limit are cited. Sources: ai-maestro:ltm.
Duplicate actions [partial] Stop-hook dedup of delivered messages; duplicate typing risks are discussed. Sources: ai-maestro:hook, ai-maestro:chatarch.
Manual repair [partial] Troubleshooting docs and the Doctor. Sources: ai-maestro:readme.
Multi day continuity [supported] Long-term memory and persistent identity target months-long continuity. Sources: ai-maestro:ltm.
Per-host installer and scripts for CLIs, Node and tmux; no snapshot or service orchestration; separate clones per agent; no automatic port allocation between workspaces beyond Docker port assignment.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: install.sh sets up the dashboard; each CLI and its dependencies are installed by the user (install-pi.sh, install-claude-vertex.sh).
caches: No cache sharing between agents.
snapshots: No snapshots; agent export/import archives exist.
services: No service orchestration for project dependencies.
10 atomic answers
Dependencies [partial] install.sh sets up the dashboard; each CLI and its dependencies are installed by the user (install-pi.sh, install-claude-vertex.sh). Sources: ai-maestro:readme, ai-maestro:programs.
No time, turn, token or spend limits; AMP is rate limited and Docker agents have CPU/memory caps; tmux and hook timeouts only.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
time limits: No run time limits.
turn limits: No turn limits.
token limits: No token limits.
spend limits: No spend limits; the cost doc explains cache effects only.
Per-agent tiles for sessions, tokens and estimated cost (Claude Code OTLP opt-in), working/idle/needs-you status from hooks and transcripts, Doctor diagnostics and conversation search/playback; no failure reasons or fleet statistics export.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
logs: Chat transcripts, terminal and server logs; log rotation script.
traces: No tracing.
failure reasons: Errors appear in chat for readers that parse them; Doctor.
usage export: Transcript export; no usage export.
13 atomic answers
Logs [partial] Chat transcripts, terminal and server logs; log rotation script. Sources: ai-maestro:programs.
Project stats [unsupported] No project statistics. Sources: ai-maestro:tasktype.
Time stats [partial] Timestamps on cards; metrics tiles lack time series. Sources: ai-maestro:taskreg.
Failure status [partial] Needs-you and offline states; no explicit failed state. Sources: ai-maestro:programs.
Approval status [supported] Needs-you status and approval cards. Sources: ai-maestro:programs.
Reconnect consistency [partial] Hooks and pane parsing reconcile state; a stale "working" file is not trusted after 10 minutes. Sources: ai-maestro:programs.
Performance method [unsupported] No benchmark method; ADEMark does not list this product. Sources: ai-maestro:readme.
Responsiveness [unknown] Not measured by this catalog. Sources: ai-maestro:readme.
Claude Code agents can show tokens and an estimated cost tile when OTLP telemetry is turned on; a script breaks down a session transcript by cache share; no per-task, epic or subscription-aware cost.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: No per-card cost.
session cost: Per-agent (Claude session) estimated cost when telemetry is enabled.
Docker CPU/memory limits per agent and a 60 s scheduler; host load hints and a bounded embedding process; no queues, backpressure or check deduplication for agent work.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
concurrency: Memory indexing bounded; scheduler skips overlapping runs; no global concurrency cap.
queues: Wake queue only.
backpressure: No backpressure.
host awareness: Host load hints (host-load.ts) and Linux sizing guide.
12 atomic answers
Concurrency [partial] Memory indexing bounded; scheduler skips overlapping runs; no global concurrency cap. Sources: ai-maestro:agentsched, ai-maestro:readme.
Not designed for several humans on one fleet ("no attribution or handoff model exists"); multiple browsers can open the same agent, agents have a people directory and calls, and cards support co-owner agents; no presence or co-prompting rules.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 device on the network sees the same dashboard, but there are no user accounts.
discussion: Card comments with operator/agent authors; operator is not identified individually.
presence: Agent presence only; no human presence.
coprompting: Several browsers can send to one agent; no ordering rules.
11 atomic answers
Visibility [partial] Any device on the network sees the same dashboard, but there are no user accounts. Sources: ai-maestro:product, ai-maestro:security.
Discussion [partial] Card comments with operator/agent authors; operator is not identified individually. Sources: ai-maestro:tasktype.
No organization/project/task roles, membership, SSO or provisioning; only agent messaging roles and a People directory.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 human roles.
project roles: No project roles.
task roles: No task roles.
membership: A People directory and agent managers exist but are not memberships with permissions.
9 atomic answers
Organization roles [unsupported] No human roles. Sources: ai-maestro:srch-sso.
Approvals for the CLIs' own prompts appear as chat cards (Claude from hook and pane; Codex, Grok, Copilot from pane); permission modes per CLI; secrets requests and SQL write requests need a person; full autonomy is selectable; tmux agents are not sandboxed.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
network: No network approval layer.
delivery: agentic-sql write requests require person approval; no merge/delivery gate.
approver identity: No approver identity.
lifetime: Each CLI's own scopes.
14 atomic answers
Command [supported] Command approval cards in chat for Claude Code, Codex, Grok and Copilot. Sources: ai-maestro:programs.
Tool [supported] Tool and edit approval cards. Sources: ai-maestro:programs.
Messages, transcripts, card comments and playback persist locally; an audit trail is only a draft spec; history is files the agent user can edit.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
actions: Transcripts and commit-to-session trace; no action log.
approvals: Approval clicks are not logged (audit trail unbuilt).
No review workflow: no diff viewer, request-changes flow or shipping guidance; chat cards, attention page and mobile layouts help triage; review is done in the terminal or Git host.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
pending reviews: Cards in review status appear on the Attention page.
change understanding: No diff UI.
evidence navigation: Commit-to-session trace.
comments: Card comments only.
13 atomic answers
Pending reviews [partial] Cards in review status appear on the Attention page. Sources: ai-maestro:tasksdoc.
Keyboard [partial] Documented shortcuts cover terminal scrolling and copy/paste only, plus search and playback keys in code; no documented keyboard flow for review or approval. Sources: ai-maestro:qs-keys.
A different person can open the same dashboard and agent chat, and agents can be transferred between hosts with their repositories; there is no attribution or handoff model.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: Any browser can drive an agent; transfer moves it to another host.
same task: Cards visible to all.
decisions: No decision log.
artifacts: Files and canvas pages stay with the agent.
Canvas pages drawn by agents, chat history, uploads, AMP attachments and shared files (AFP) persist; no diff, screenshot or recording gallery and no test/reviewer outputs.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
diffs: No diff artifact.
screenshots: Attachments can carry images; no gallery.
No verification gates: tests, lint, CI and review are whatever the agent runs; no required checks or spec checks.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · Plan: Open source (MIT) · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Agents run git themselves in their own clones; the product reads repos and traces commits to sessions; Docker/EC2 agents receive a GitHub token; no PR creation or CI response flow.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: Per-agent clone with own branch by convention.
worktrees: Not used by the product.
commits: Commits traced to sessions.
pull requests: No PR code found.
9 atomic answers
Branches [partial] Per-agent clone with own branch by convention. Sources: ai-maestro:readme.
No merge, release or deploy authority in the product; delivery happens through git by agents and humans, with no revision-bound approvals.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 approval flow.
merge: No merge flow.
release: None.
deploy: Dashboard deploys agents to EC2/ECS, not user code.
Commit-to-session tracing only; no PR, build or deployment links, delivery status, rollback or stale-approval handling.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: GET /api/commits/:sha finds the agent session (Claude Code only).
No review queue or batch delivery features; Attention page lists cards in review and overdue work.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: Cards with status review appear on the Attention page.
summaries: None.
grouping: Matrix grouping by quadrant and Org view.
risk size filters: None.
12 atomic answers
Review queue [partial] Cards with status review appear on the Attention page. Sources: ai-maestro:tasksdoc.
MIT for the dashboard and server (LICENSE, package.json); plugin skills and gateways are separate repositories; the iPhone and Android apps are closed; pinned v0.106.98 (3acb573).
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 extensions: No hosted service exists.
source visibility: Dashboard, plugin builder and gateways are public; mobile apps are not found in public repos.
Hosted extensions [unsupported] No hosted service exists. Sources: ai-maestro:readme.
Source visibility [partial] Dashboard, plugin builder and gateways are public; mobile apps are not found in public repos. Sources: ai-maestro:srch-clients, ai-maestro:gateways.
Fully self-hosted with no vendor service; state is JSON/SQLite/CozoDB files under ~/.aimaestro; agents can be exported and imported as archives and transferred between hosts; transcript export exists.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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: Version check hits raw.githubusercontent.com; optional TTS and Expo push use third parties; the iPhone app is distributed by Apple.
formats: JSON files and archives.
cancellation: No account exists to cancel.
history portability: Transcripts export; memory graph is CozoDB.
8 atomic answers
Self hosting [supported] Runs entirely on user hosts, with headless workers. Sources: ai-maestro:readme.
Vendor dependencies [partial] Version check hits raw.githubusercontent.com; optional TTS and Expo push use third parties; the iPhone app is distributed by Apple. Sources: ai-maestro:versioncheck, ai-maestro:pushexpo.
Export [supported] Export agent and transcript export dialogs. Sources: ai-maestro:readme.
Import [supported] Import agent dialog and aimaestro-task.sh import. Sources: ai-maestro:tasksdoc.
Everything stays on your hosts; the README states no telemetry, no account and no phone-home; Claude telemetry export is opt-in to your own receiver; optional TTS and push call third parties; the iPhone app sends crash reports unless disabled.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
Version check is not documented as controllable.
15 atomic answers
Code data [supported] Code stays on user machines; model providers see what the CLI sends. Sources: ai-maestro:readme.
Prompts [supported] Prompts go to the user CLI's provider, not to AI Maestro servers. Sources: ai-maestro:readme.
Free and MIT with no seats or usage fees; optional iPhone app is free for 30 days then $19.99 one-time (USD); agent subscriptions and your cloud costs are separate.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · Plan: Free MIT; optional iPhone app · 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
included usage: No metered usage.
credits: No credits.
overages: None.
taxes: Tax treatment on the App Store purchase is not stated.
12 atomic answers
Currency [supported] USD for the iPhone app ($19.99). Sources: ai-maestro:iphonepage.
Interval [supported] One-time purchase for the app; nothing recurring. Sources: ai-maestro:iphonepage.
Scenario costs are not priced by the vendor; the ADE is free, so only CLI subscriptions, hardware or cloud and the optional $19.99 app apply; no invented averages.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
one person: ADE $0 plus the user's agent subscriptions and optional $19.99 app.
two people: No multi-user pricing; each person would use their own hosts.
Generally available open-source releases at a high cadence (v0.106.98 on 2026-11-07); support is through GitHub issues; no response commitments or status page found; iPhone app on the App Store, Android in invite beta.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
access restrictions: Android is an email-invited beta.
support: GitHub issues and support email for app testers.
response commitments: None stated.
status history: No status page.
8 atomic answers
Release stage [supported] Release v0.106.98 and App Store version 1.0.5; version numbers below 1.0 for the server. Sources: ai-maestro:release, ai-maestro:appstore.
Access restrictions [partial] Android is an email-invited beta. Sources: ai-maestro:iphonepage.
No ISO or SOC 2 claims are published for this MIT project; SECURITY.md lists advisories and the network trust model.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
iso scope: No ISO certificate claim found in the repository, SECURITY.md or website.
iso issuer: No public certification statement or report found in the checked sources.
iso validity: No public certification statement or report found in the checked sources.
soc2 type: No SOC 2 report claim found in the repository, SECURITY.md or website.
9 atomic answers
Iso scope [unknown] No ISO certificate claim found in the repository, SECURITY.md or website. Sources: ai-maestro:security.
Iso issuer [unknown] No public certification statement or report found in the checked sources. Sources: ai-maestro:security.
Iso validity [unknown] No public certification statement or report found in the checked sources. Sources: ai-maestro:security.
Soc2 type [unknown] No SOC 2 report claim found in the repository, SECURITY.md or website. Sources: ai-maestro:security.
Soc2 scope [unknown] No public certification statement or report found in the checked sources. Sources: ai-maestro:security.
Soc2 period [unknown] No public certification statement or report found in the checked sources. Sources: ai-maestro:security.
Report access [unknown] No public certification statement or report found in the checked sources. Sources: ai-maestro:security.
Verification date [unknown] No public certification statement or report found in the checked sources. Sources: ai-maestro:security.
Evidence status [partial] Public advisories and a security policy exist; no attestations. Sources: ai-maestro:security.
Free to use in full as MIT software with no card; iPhone app has a 30-day trial then $19.99 one-time; Android beta by invitation.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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 trial: None.
paid only: No paid-only core.
early access: Android beta by invitation.
free scenarios: Everything except mobile apps works free; iPhone is a free trial.
10 atomic answers
Ongoing free [supported] Whole product is free. Sources: ai-maestro:readme.
Transparent: free MIT software, no seats or usage fees; the only price is the one-time $19.99 iPhone app after a 30-day trial; model and compute costs are the user's.
Scope
Surfaces: Web dashboard and headless server (localhost:23000, tmux agents) · Version: v0.106.98 (commit 3acb573, released 2026-11-07) · 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
overages: None.
compute markup: No markup; user pays cloud directly.
MIT dashboard for tmux-hosted coding agents across a peer mesh; Claude Code/Codex/Grok Build first-class, Pi/OpenClaw/Copilot CLI launch only; Windows via WSL2; no telemetry; no hosted tier; iPhone app listed as coming soon in this README.
Per-program support: Claude Code, Codex CLI, Grok Build get launch, chat history, status and approval cards; Pi, Copilot CLI, OpenClaw recognised with limits; Cursor, Gemini CLI, OpenCode have no reader; Aider deprecated.
iPhone app on the App Store (30 days free, then $19.99 one-time) lets you see status, answer permission prompts and questions, chat, open a terminal and start agents; Android 7+ is a beta joined by emailing support; the app needs AI Maestro running on a computer you control.
Kind
documented
Locator
schema.org MobileApplication; "When can I get it?"; Android answer
Binds 0.0.0.0:23000 with no application-level authentication by design (network is the perimeter); tmux sessions persist after the dashboard closes and run as the user; two tmux command-injection CVEs fixed.
Kind
documented
Locator
Network Access Configuration; By design no authentication; tmux Session Security; Resolved Advisories
Peer mesh with no central server; each agent is a tmux session plus working directory; agents own repositories and are not given shared working copies.
Kind
documented
Locator
Peer Mesh Network; Agents and tmux Sessions; Agent Ownership
Primary user is a solo operator; "Not currently designed for multiple humans sharing one fleet (no attribution or handoff model exists)"; no authentication by design; responsive desktop, tablet and phone layouts.
Cards on agent, team and project boards with status, blockedBy, quadrant, kind (task/idea/bug/research), origin, due, owners; Matrix, Org and Attention pages; agent CLI and REST API; commit-to-session trace for Claude Code sessions only.
Kind
documented
Locator
What a card holds; For agents: aimaestro-task.sh; API; Commit to session
No Electron, Tauri, VS Code or JetBrains extension; only a web app, PWA manifest and Expo push; the iOS and Android apps are not in the public repositories.
Kind
source-inspected
Locator
grep -Ei over app, lib, services, components, package.json for electron, tauri, vsix, jetbrains plugin, vscode extension
Default tmux chat path types into the terminal and reads the screen; the Streaming tab uses the Claude Agent SDK with structured events and canUseTool approvals.
One hook script for Claude Code (Stop, Notification idle_prompt, SessionStart, UserPromptSubmit), Codex (Stop, SessionStart), Gemini and Grok; delivers unread AMP messages and reports working/idle/needs-you status.
Docker agents run in a container with dropped capabilities, no-new-privileges, noexec tmpfs, CPU and memory limits (default 2 CPUs, 4g) and a bind-mounted workspace.
An MCP server (claude/channel capability, stdio) that Claude Code spawns to push AMP inbox events into a running session; registers a localhost port under ~/.aimaestro/channels.
Bounded grep finds no API stability or deprecation policy for the dashboard REST API; package.json is 0.106.x and only the AMP routes carry an /api/v1 prefix.
aim-secret vault (Keychain, libsecret or encrypted file); agent uses secrets by name with values scrubbed; file backend protects against accidents only.
blockedBy is stored on each card; isBlocked and blocks are derived; cycles are refused; completing a card returns newly unblocked cards. No dispatch gate on blockers found.
A card assigned to an agent can be forwarded as a typed instruction to the assignee agent session on another host (POST /api/agents/:id/task-instruction).
The Matrix page filters cards by agent name or alias; no free-text search over card subjects or descriptions was found in app/matrix, components/matrix or AgentBoardPanel.
Tasks are read-modify-written as a JSON file synchronously in one process; the grep found no lock, version check or merge step, so concurrent writers would overwrite each other.
Kind
source-inspected
Locator
writeFileSync at line 71; grep for lock|mutex|etag|expectedVersion
Off by default: AIMAESTRO_TELEMETRY opts Claude Code launches into exporting OTLP metrics to the local AI Maestro receiver (or a central collector the operator sets).