50 of 51 criteria have a sourced finding; 43 were also cross-checked by a second agent. Every comparison that includes no_human 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.
Desktop app (macOS arm64 DMG, Windows unsigned exe, Linux .deb and AppImage x86_64), a web board served on loopback, the nh CLI with a shell, an MCP stdio bridge and a Claude Code plugin. No mobile apps; the board has no authentication so phone access is not a supported path.
Scope
Surfaces: no_human desktop app (Electron shell around the local nh server), Web board on 127.0.0.1:8420, nh CLI and shell, MCP stdio bridge, Claude Code plugin and GitHub Action reviewer gate · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
9 atomic answers
Desktop os [supported] macOS (arm64 DMG), Windows unsigned installer and Linux .deb/AppImage ship per release; windows and linux docs mark parts unverified. Sources: no-human:release, no-human:windows, no-human:linux.
Mobile apps [unsupported] No iOS or Android app in desktop, web or plugins. Sources: no-human:srch-mobile.
Browser [supported] nh start serves the board and worker on 127.0.0.1:8420. Sources: no-human:readme, no-human:api.
Mobile browser [partial] server.host can bind another address, but the board API is unauthenticated and loopback-checked; no phone workflow is documented. Sources: no-human:boundary, no-human:config.
Cli [supported] nh with shell, task, status, review, diff, approve, reject, reply, wake and more. Sources: no-human:cli, no-human:readme.
Editor extensions [partial] A Claude Code plugin and MCP bridge file tasks from the editor session; no VS Code or JetBrains extension. Sources: no-human:plugin, no-human:srch-mobile.
Task actions [supported] Add, pause, resume, cancel, retry, answer questions, review, diff, approve, reject and send back from board or CLI. Sources: no-human:api, no-human:cli.
Notifications [partial] Slack and Teams messages when a task needs you plus a Needs answer lane; no push to a phone. Sources: no-human:readme, no-human:adapters.
Cross device continuity [unsupported] State is a local SQLite database on one machine. Sources: no-human:site, no-human:srch-cloud.
Three coder backends: claude (Claude Agent SDK driving the claude CLI, default), codex (codex exec) and local (the claude CLI pointed at your own Anthropic-compatible server). Planner, supervisor, intake and by default the reviewer stay on Claude; the reviewer can be assigned a different backend per role, so a codex coder with a Claude reviewer is a mixed-harness run.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
5 atomic answers
Harnesses [supported] claude and codex CLIs plus a local-server mode of the claude CLI. Sources: no-human:backends.
Model providers [partial] Anthropic, OpenAI and any Anthropic-compatible server; the model picker lists enabled models; other providers need the local mode. Sources: no-human:backends, no-human:config.
Custom adapters [unsupported] The backend list is closed to three entries; the local mode is the only extension point. Sources: no-human:backends.
Session integration [supported] Embeds the Claude Agent SDK and runs codex exec --json with per-task preflight. Sources: no-human:backends, no-human:claude.
Mixed harness operation [supported] A codex coder with a Claude reviewer bills two vendors in one task. Sources: no-human:backends.
Claude subscription OAuth (claude setup-token) or Anthropic API key; Codex through an OpenAI API key or a ChatGPT login (codex_auth_mode subscription); local Anthropic-compatible servers. One billing path per run is enforced by scrubbing other credentials. No vendor credits; no account required.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
7 atomic answers
Agent subscriptions [supported] Claude OAuth token in subscription mode (default) and Codex ChatGPT subscription mode. Sources: no-human:security, no-human:backends.
Vendor credits [unsupported] No vendor credits; the local tier is free and inference is billed by the user's provider. Sources: no-human:site.
Local models [supported] llm.local_model and llm.local_base_url point the claude CLI at your own server. Sources: no-human:config, no-human:backends.
Account switching [partial] nh auth use switches the auth mode and backend; one credential per run. Sources: no-human:cli, no-human:security.
Vendor account requirement [supported] No no_human account; official builds forward the wizard email to a hosted registration intake. Sources: no-human:site, no-human:email.
Provider restrictions [partial] INTEGRATIONS_LEGAL.md records the legal position and unresolved points for Codex authentication and other integrations. Sources: no-human:legal.
Quickstart in five minutes: install prerequisites, nh init (token, config, first repo), add a task from a title or a GitHub/GitLab/Jira/Linear/monday ticket, run, review, approve. Needs a git repository; PR opening and landing use a GitHub or GitLab remote. No managed cloud or VPS installer.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
7 atomic answers
First task setup [supported] Eight-step quickstart with nh doctor to prove the install. Sources: no-human:quickstart.
Local git without remote [partial] Worktrees and branches work locally; PR opening and landing need a git host. Sources: no-human:quickstart, no-human:config.
Hosted git [supported] GitHub and GitLab issues import by URL and PRs/MRs open on your host. Sources: no-human:readme, no-human:adapters.
Prerequisite guidance [supported] Python 3.12+, uv, git, Node for the board build from source, tokens and nh doctor. Sources: no-human:readme, no-human:quickstart.
Managed cloud setup [unsupported] No managed cloud exists; Max cloud is waitlist-only. Sources: no-human:site, no-human:srch-cloud.
Byo vps onboarding [unsupported] No remote or VPS installer in the release. Sources: no-human:srch-cloud.
nh CLI, a local HTTP API behind the board (tasks, approvals, events, metrics, config), an MCP stdio bridge (task_add, task_status), a Claude Code plugin and a GitHub Action reviewer gate. Board API has no authentication beyond loopback checks; no SDK or inbound webhooks.
Scope
Surfaces: nh CLI and shell, Web board on 127.0.0.1:8420, MCP stdio bridge, Claude Code plugin and GitHub Action reviewer gate · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
8 atomic answers
Cli [supported] Click CLI with many command groups. Sources: no-human:cli.
Api [partial] Local /api routes for tasks, approvals, events and config; unauthenticated and not documented as a public API. Sources: no-human:api, no-human:boundary.
Sdk [unsupported] No SDK in the repository. Sources: no-human:srch-spec.
Mcp server [supported] nh mcp-serve exposes task_add and task_status over stdio; listed in the MCP registry. Sources: no-human:mcp-server, no-human:plugin.
Webhooks [partial] Slack and Teams notifications go out; no inbound webhook triggers; trackers are polled. Sources: no-human:readme, no-human:adapters.
Authentication [partial] Loopback Host and Origin checks; the Action takes a credential input; MCP bridge uses the local server. Sources: no-human:boundary, no-human:action.
Stability [partial] Pre-1.0 (0.2.4) with Keep a Changelog entries; no API stability promise found. Sources: no-human:changelog.
Intake integrations (Jira, Linear, monday, GitHub, GitLab) with write-back, Slack and Teams notices, CI backends, a Claude Code plugin and the MCP bridge. The coder reads project settings and skills through the Claude Agent SDK; no general MCP client configuration, plugin marketplace or custom tool API of its own.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
7 atomic answers
Mcp client [partial] The coder session loads project settings so MCP servers configured in the repository are available; no no_human MCP configuration surface. Sources: no-human:claude.
Plugins [partial] Claude Code plugin and nh skills (nh skills add) exist; no third-party plugin API. Sources: no-human:plugin, no-human:cli.
Custom tools [partial] Verifiers and playbooks add checks and rules; no custom tool registry. Sources: no-human:verification, no-human:cli.
Instructions import [supported] Repository .no_human.yml and project.yml plus nh rules and skills. Sources: no-human:config, no-human:cli.
Authentication [supported] Tokens stored in ~/.no_human/.env (0600) and never in config or repo. Sources: no-human:security, no-human:config.
Permission scope [partial] Coder runs bypassPermissions by default with acceptEdits optional; tracker filters live in config. Sources: no-human:security, no-human:readme.
Integration owner [supported] no_human code owns intake, notices and CI adapters; the Claude CLI owns MCP. Sources: no-human:adapters.
Electron desktop shell around a Python server and a web board built with npm; the desktop attaches to a local nh server or spawns the frozen one. No mobile client.
Scope
Surfaces: no_human desktop app (Electron shell around the local nh server), Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Tasks start from a freeform title or sentence, a GitHub or GitLab issue URL, or Jira, Linear and monday.com tickets polled via filters; each ticket is scoped with the user (intake grill). With write-back on, ticket status and the PR link go back to the tracker. The tracker stays authoritative for tickets; no_human keeps task and attempt records.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
9 atomic answers
Local creation [supported] nh task add with a title or sentence works with no tracker. Sources: no-human:quickstart.
Prompt intake [supported] Plain-English intake in the shell and board. Sources: no-human:readme.
A plan is written before any code from the ticket and repo (understanding as acceptance criteria, files, approach, test plan, scope, verification command); the intake grill asks clarifying questions; plan skip is stated in the event stream; per-attempt resume checkpoints exist. No plan editor with revisions.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
7 atomic answers
Plans [supported] Plan with acceptance criteria, files and verification command before coding. Sources: no-human:readme.
Acceptance criteria [supported] Checkable acceptance criteria in the plan; reviewer ticks each with file and line. Sources: no-human:readme, no-human:verification.
Editors [partial] Answers and send-back feedback steer the plan; no direct plan editor. Sources: no-human:api, no-human:cli.
Revisions [partial] Attempts and events are retained; no plan version history. Sources: no-human:api.
Implementation links [supported] Reviewer checklist cites file and line per criterion. Sources: no-human:readme.
Amendment [partial] Pause, send back or reply with an answer; a running attempt checkpoints. Sources: no-human:api, no-human:config.
Tasks can be split into subtasks (split proposals), follow another task (--follows), and wait on dependencies with DEPENDENCY_WAIT blockers and merge-stack link, plan and run for stacked PRs; priority orders the queue; no epics or free-depth hierarchy.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Hierarchy [partial] Tasks split into subtasks and merge stacks link PRs; one level of subtasks. Sources: no-human:api, no-human:cli.
Blockers [supported] Blocker taxonomy including DEPENDENCY_WAIT with wake conditions. Sources: no-human:blockers.
Readiness [partial] Needs-you, waiting, working and done lanes; blockers carry wake conditions. Sources: no-human:blockers, no-human:readme.
Enforcement [partial] Blocked tasks wait on a wake condition; merge-stack runs in order on the user's command. Sources: no-human:blockers, no-human:cli.
State propagation [partial] Supersession warning on approve; no rollup. Sources: no-human:changelog.
Identity retention [partial] Split drafts create subtasks linked to the parent. Sources: no-human:api.
Workflow constraints [supported] The plan, implement, review pipeline is fixed but configurable by tier and routing. Sources: no-human:verification, no-human:config.
Repository config (.no_human.yml, project.yml), nh rules, playbooks, skills, memories (learnings with confirm, reject, retire and a 45-day archive sweep) and a wiki job feed later tasks; quarantine for suspect memories. Shared team brain is an announced Pro item.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
A repository map, wiring evidence (is the symbol reachable), lint and net-new type diagnostics feed the reviewer; a brain/context layer indexes the repo; no external knowledge connectors beyond trackers; attachments accepted on tasks.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
10 atomic answers
Code search [supported] git grep based wiring evidence and exploration tools in the coder session. Sources: no-human:verification.
Native board with lanes Needs answer, Working, Review PR, Backlog, Done, Failed; task priority high/medium/low, search, retitle, pause, resume, cancel, retry; tracker backlog picker; no labels or assignee fields. Tasks and attempts are separate entities.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
17 atomic answers
Create [supported] nh task add, board New task and tracker picks. Sources: no-human:readme.
Edit [partial] Retitle and config; plan edits by feedback. Sources: no-human:cli.
Discuss [partial] Reply and send-back threads with the agent; single operator. Sources: no-human:api.
Agents (the coder) work inside a task, ask a specific question that parks the task (Needs answer), create no issues of their own; the MCP bridge and Claude Code plugin let another agent add and check tasks. Task IDs are shared by board, CLI, API and MCP; subtask split proposals and blockers are system-created.
Scope
Surfaces: MCP stdio bridge, Claude Code plugin and GitHub Action reviewer gate, Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
No living product specification: acceptance criteria belong to each task plan and memories are rules, not requirements; no revisions, drift detection or spec review gate.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
Persistent requirements [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Human edits [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Agent edits [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Revisions [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Approvals [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Implementation links [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Later retrieval [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Drift detection [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Review integration [unsupported] No persistent requirements object, approvals, drift detection or spec review integration found in docs or src. Sources: no-human:srch-spec.
Concurrency is off by default; when enabled a worker pool (default 2 workers) runs tasks in parallel worktrees; the header shows busy workers and queued count; lanes show needs-you, working, review and done. Overnight drain runs a bounded pool.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
5 atomic answers
Concurrent sessions [supported] Parallel task workers, each in its own worktree. Sources: no-human:config.
A queue ordered by high/medium/low priority dispatches pending tasks to workers; quota-parked resumes first; leases prevent two workers owning a task (lease takeover after 600 s); no human assignment.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
7 atomic answers
Human assignment [unsupported] Single operator, no assignment. Sources: no-human:srch-team.
Agent assignment [partial] Dispatcher assigns tasks to workers; no choice of agent per task beyond --backend. Sources: no-human:config, no-human:backends.
Claims [supported] Worker leases with takeover and orphan grace. Sources: no-human:config.
Queues [supported] Pending queue with priority and no aging. Sources: no-human:config.
Product-owned pipeline of roles: planner, coder, an independent fresh-context reviewer (different model, never saw the coder transcript), supervisor, utility and intake agents; results return as plan, checklist and PR body; model per tier is configurable; the user does not invoke subagents directly. Cross-vendor coder and reviewer possible.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Model harness effort [supported] llm.primary_model, review_model, codex model and reasoning effort, model picker. Sources: no-human:config, no-human:backends.
Messages [partial] Send-back feedback and reply; roles do not converse freely. Sources: no-human:api.
No built-in scheduler: nh watch and the overnight drain (--until-empty from cron or CI) run pools, trackers are polled on a tick, a nightly eval and a wake watcher resume blocked tasks; GitHub Action runs on PR events. Runs on the local machine, so a closed laptop stops them.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
8 atomic answers
Schedules [partial] Cron or CI invokes the drain; no in-product cron. Sources: no-human:quickstart.
Pause, resume, cancel, retry, send-back, reply to a parked question, wake blockers and graceful stop with checkpoints; Slack and Teams notices when a task needs you; Needs answer lane and oldest-waiting counter; honest stop with a specific question.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
The board shows lanes across tasks with counts, oldest-waiting time, per-task attempts, tokens and elapsed time, a Stats page and last-24h shipped totals; blockers show the wake condition; no epics or dependency graph.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
8 atomic answers
Fleet fixture [supported] Board hero image shows five lanes across several tasks. Sources: no-human:readme.
Parent child [partial] Subtasks link to parents. Sources: no-human:api.
The pipeline is fixed but tunable: tiers, review routing by diff size, backend per task, repro-gate mode, verifiers per repo; natural-language intake in the shell; per-role models; changes mid-run by pause/send-back; the event stream shows what ran.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Runs on the user's machine (desktop or nh serve): coder sessions use the local claude or codex CLI in per-task worktrees. No vendor cloud, no remote worker, no VPS mode in the release; Max cloud is waitlist-only. A closed laptop stops it.
Scope
Surfaces: no_human desktop app (Electron shell around the local nh server), Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (MIT) · OS: Not specified · Execution modes: local · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-16 (6 days old)
Agent cross-check
2026-11-17
See atomic answers for the unmet or unknown parts.
Every task runs in a throwaway git worktree by default; a task can opt out and use the primary checkout. No container, VM or OS sandbox: the coder runs bypassPermissions by default, with acceptEdits optional and a codex sandbox network switch; credentials are scrubbed so one billing path is used.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
12 atomic answers
Checkout [supported] isolation.enabled false uses the primary checkout. Sources: no-human:config.
Branch [supported] Per-task branch and commits under a distinct identity. Sources: no-human:security.
Worktree [supported] git worktree per task under ~/.no_human/worktrees. Sources: no-human:config.
Container [partial] UNTRUSTED_PR_REVIEW describes reviewing external PRs in a container; the coder is not containerised. Sources: no-human:security.
Attempts checkpoint with [WIP-PARTIAL] commits and resume_from; nh stop waits a grace period; leases are taken over after 600 s; quota-parked tasks resume first; retry and restore-approval commands; orphan grace and a task-sleep style wake watcher. All state is local SQLite.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
12 atomic answers
Client disconnect [supported] The server keeps running without the window. Sources: no-human:desktop.
Worker crash [supported] Lease takeover after 600 s and orphaned task rows handled. Sources: no-human:config.
Service restart [supported] Graceful stop checkpoints attempts and resumes them. Sources: no-human:config.
setup_cmds per repo prepare a fresh task worktree (installs, builds); per-repo test, lint and type commands; no snapshots, services or port management beyond configured verifiers; no shared dependency cache.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
10 atomic answers
Dependencies [supported] setup_cmds run in each fresh worktree. Sources: no-human:config.
Caches [unsupported] No shared cache found. Sources: no-human:srch-spec.
Bounded loop: 3 attempts, 500 turns per attempt, inactivity timeout 3600 s, reviewer timeouts, min viable attempt token floor, per-task budget cap set before it runs, convergence early-abort; honest stop with a structured budget record; account quota parking. No dollar cap.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
8 atomic answers
Time limits [supported] attempt_timeout_s 3600 inactivity, review timeouts, approve test timeouts. Sources: no-human:config.
Turn limits [supported] bounds.max_turns_per_attempt 500 and convergence abort. Sources: no-human:config.
Token limits [supported] Per-task weighted token budget and a floor for starting an attempt. Sources: no-human:config, no-human:site.
Spend limits [partial] Token budgets approximate spend; no dollar cap and Enterprise budget policy is announced. Sources: no-human:config, no-human:site.
Timeouts [supported] Several documented timeouts. Sources: no-human:config.
Retries [supported] Up to three implement/review cycles plus provider retry. Sources: no-human:config.
Provider failures [supported] Quota parking and honest stop with reason. Sources: no-human:verification.
Enforcement [supported] Enforced stops with recorded reasons. Sources: no-human:verification.
Event stream per task, attempt details with failure reasons, reviewer checklist, metrics endpoints, Stats page, usage ledger, last-24h shipped and token totals, pr-outcomes and a nightly eval; anonymous telemetry separate. Speed and memory figures come only from ADEMark, which has no no_human entry.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
13 atomic answers
Logs [supported] Events stream and attempt details. Sources: no-human:api.
Traces [partial] Event log of planner, coder, review steps; no distributed traces. Sources: no-human:api.
Per-task token counts and estimated dollar cost (for example ~$0.40 per PR estimate and last-24h totals), usage ledger with 90-day compaction, cost levers analysis, weighted tokens per attempt; subscription mode shows tokens not dollars charged; no budget alert system.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
14 atomic answers
Task cost [supported] Attempts show tokens and the card shows estimated cost. Sources: no-human:readme.
Session cost [supported] Per attempt token totals. Sources: no-human:readme.
Child cost [partial] Subtasks are separate tasks. Sources: no-human:api.
Worker pool with a default of 2 workers clamped to a CPU-derived ceiling (cpu_count // 3, minimum 2), pending queue and priority, per-task worktrees, per-task pytest-xdist worker bounds, review depth scaled by diff size, change-scoped test runs and full-suite fallback with timeouts, minimum viable attempt token floor; no RAM limits, live load awareness or check reuse across tasks.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Cpu ram [partial] CPU only: pool width ceiling is cpu_count // 3 (minimum 2) and xdist auto workers per task are bounded to cpu // max_workers; no memory limit. Sources: no-human:sched-cpu.
Backpressure [partial] Quota parking and floor token gate before starting an attempt. Sources: no-human:config.
Host awareness [partial] The detected CPU count sets the pool ceiling at server start and Settings shows it; no live memory or load awareness found. Sources: no-human:sched-cpu.
Check reuse [partial] Change-scoped test runs avoid the full suite; no cache across tasks. Sources: no-human:config.
Check deduplication [unsupported] No deduplication across tasks. Sources: no-human:srch-spec.
Bounded tests [supported] approve_merge test timeouts 1800 s and 5400 s. Sources: no-human:config.
Single-operator local app: no presence, comments between people or co-prompting; Slack and Teams notices go to a channel; the board API is unauthenticated on loopback.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
11 atomic answers
Visibility [unsupported] No shared visibility of the local board. Sources: no-human:srch-team.
Discussion [unsupported] None between people. Sources: no-human:srch-team.
Comments [partial] Reviewer checklist is posted as a PR comment; no in-app comments. Sources: no-human:config.
No roles, SSO or provisioning in the local product; Enterprise tier lists SSO and SCIM (not released). Board access is loopback-only.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
Organization roles [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Project roles [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Task roles [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Membership [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Sso [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Provisioning [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Unauthorized access [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Revocation [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
Plan scope [unsupported] Bounded search found no accounts, roles, SSO or provisioning in the release; SSO and SCIM appear only on the waitlisted Enterprise tier. Sources: no-human:srch-team, no-human:site.
The agent never merges: nh approve is the human action that lands a PR under git.approve_identity, with change-scoped test gating; coder permission mode is bypassPermissions by default or acceptEdits; reviewer and tamper guard are enforced; Merge-ready policy. No network or secret approval flows.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
14 atomic answers
Command [partial] Coder runs with bypassPermissions by default; acceptEdits optional. Sources: no-human:security.
Tool [partial] Per-tool prompts are skipped by default. Sources: no-human:security.
Append-only event log per task, attempt records, reviewer checklist comment on the PR, usage ledger, signed distinct git identity; local SQLite so an operator can alter it; no audit retention or integrity model (an audit trail is Enterprise-announced).
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
11 atomic answers
Actions [supported] Events per task. Sources: no-human:api.
Approvals [supported] Approval and reject events with reasons. Sources: no-human:cli, no-human:api.
Edits [supported] Commits and diffs under a distinct identity. Sources: no-human:security.
Discussion [partial] Replies and send-back feedback. Sources: no-human:api.
Decisions [partial] Reviewer verdicts and approval records. Sources: no-human:verification.
Cross user persistence [unsupported] Single operator. Sources: no-human:srch-team.
Review lane with the reviewer's evidence checklist (pass/fail with file and line), diff view, test results panel, tamper banner, reproduction gate verdict, plan with acceptance criteria, PR body sections the coder cannot author, and one-click approve or reject with reason; PR comment of the checklist; CLI review, diff and review-comments.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Change understanding [supported] Plan, checklist and diff. Sources: no-human:readme.
Evidence navigation [supported] Test results, events, tamper and repro verdicts per task. Sources: no-human:readme.
Comments [partial] Reviewer comment on the PR; review-comments command. Sources: no-human:config, no-human:cli.
Request changes [supported] Reject with reason or send back. Sources: no-human:cli.
Shipping guidance [supported] Review and approve step with NOT RUN clarity. Sources: no-human:readme.
Effort [unknown] Review effort is a usability measure outside source inspection. Sources: no-human:readme.
Missed evidence [supported] NOT RUN labelled on PR when no test command is found. Sources: no-human:readme.
Keyboard [partial] Task rows and cards open with Enter or Space, dialogs trap Tab focus and Escape closes slide-overs; no shortcut list or command palette found. Sources: no-human:web-kbd.
Narrow screen [supported] The web board has a phone layout at a 640px breakpoint shared by JavaScript and CSS; it is served on loopback, so a phone needs a separate unauthenticated bind. Sources: no-human:web-resp.
Context switching [partial] One task at a time; lanes list tasks. Sources: no-human:readme.
Single-operator: no colleague takeover; the work and history live in one local database; the PR and checklist comment are the shared artefacts for colleagues; the Pro shared brain is waitlisted.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
12 atomic answers
Ongoing takeover [unsupported] No second-user access. Sources: no-human:srch-team.
Plan, diff, reviewer checklist, test results panel, tamper and repro-gate verdicts, event log, PR body with evidence sections and optional UI evidence (screenshots) per repo config; retained per task; no cross-task gallery.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Required by default: tests and lint (your commands), wiring and type evidence as inputs, tamper guard before review, reproduction gate (Python bug fixes by default), independent adversarial review, merge-ready policy and change-scoped tests at landing; CI backends (GitLab CI, GitHub Actions, Jenkins, CircleCI) optionally gate; failed checks block approval and landing. No persistent-spec checks.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
11 atomic answers
Tests [supported] Your tests run locally and optionally through CI; NOT RUN is stated. Sources: no-human:readme.
Lint [supported] Deterministic lint evidence as reviewer input. Sources: no-human:verification.
Types [supported] Net-new type diagnostics as input. Sources: no-human:verification.
Integrated: per-task worktrees and branches, deterministic VCS operations under a distinct identity, lint-gated commits, PR/MR opening on GitHub or GitLab, reviewer checklist comment, CI response through backends, squash landing and merge-stack for stacked PRs; supersession warnings.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
The human approves with nh approve or the board: approval squashes the branch into one commit, runs change-scoped tests, pushes the default branch and closes the PR (can be disabled to leave the merge to you); the agent never merges; merge-stack run calls gh pr merge on your command; no deploy or release.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Exact revision [supported] Landing compares the squash tree with the tested tree and re-runs the full suite when they differ. Sources: no-human:config.
Authority enforcement [supported] The agent merge ban is enforced in code. Sources: no-human:security.
Task rows link PR, status and verdicts; shipped endpoint records delivery; attempt records bind the tested tree; PR outcomes can be refreshed; superseded tasks warn on approve; no rollback or revert feature; approval records can be restored.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
9 atomic answers
Commit links [supported] Attempts record commits and tested tree. Sources: no-human:config.
Pr links [supported] PR link on the task card. Sources: no-human:readme.
Build links [partial] CI backend results are attached. Sources: no-human:adapters.
Review PR lane with checklists, nh status and review commands, merge-stack plan and run for dependent PRs, overnight drain to open PRs for review in the morning; no risk or size filters beyond review routing; stale approvals are re-tested at landing.
Scope
Surfaces: Web board on 127.0.0.1:8420, nh CLI and shell · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
12 atomic answers
Review queue [supported] Review PR lane and nh status. Sources: no-human:readme.
Summaries [supported] Plan, checklist and PR body summaries. Sources: no-human:readme.
Grouping [partial] Lanes group by state; stacks group PRs. Sources: no-human:cli.
Risk size filters [partial] Review routing treats risk-flagged diffs separately; no user filter. Sources: no-human:config.
Findings [supported] Retained checklists per attempt. Sources: no-human:api.
Artifacts [partial] Plan, test results and diff. Sources: no-human:readme.
Request changes [supported] Reject and send back. Sources: no-human:cli.
Exact revision approval [supported] Tested-tree comparison at landing. Sources: no-human:config.
Dependency queues [supported] merge-stack link, plan and run. Sources: no-human:cli.
Next actions [supported] Review and approve prompts, NOT RUN labels, Needs answer. Sources: no-human:readme.
Stale approval invalidation [supported] Tests re-run on a diverging tree. Sources: no-human:config.
Check deduplication [partial] Change-scoped tests; no cross-task dedupe. Sources: no-human:config.
MIT License for the whole repository (CLI, server, web board, desktop shell, plugin and Action); one public repository; official builds stamp a hosted registration endpoint for the onboarding email only. Pinned v0.2.4.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
6 atomic answers
Client license [supported] MIT for web board and desktop shell. Sources: no-human:license.
Hosted extensions [partial] Hosted registration intake exists for onboarding emails and Max cloud is planned; their code is not in the repository. Sources: no-human:email, no-human:site.
Source visibility [supported] Public repository no-human-ai/no_human. Sources: no-human:repoapi.
Source revision [supported] v0.2.4 at commit a00adde. Sources: no-human:release.
Hosted parity [unknown] Parity with Max cloud cannot be established; it is unreleased. Sources: no-human:site.
Local-first: SQLite state and ~/.no_human config, no required vendor service, MCP and Action standalone; exports are the API and the git PRs; official builds forward only the onboarding email. Desktop updates via GitHub Releases; wheel on PyPI.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
8 atomic answers
Self hosting [supported] Runs entirely on user hardware. Sources: no-human:site.
Vendor dependencies [partial] Onboarding email endpoint in official builds, PostHog telemetry (opt-out), a desktop update check against GitHub Releases that no setting disables, and model providers; none is required to run the server. Sources: no-human:email, no-human:telemetry, no-human:updater.
Export [partial] API and PRs; no bulk export. Sources: no-human:api.
Import [partial] Tracker and issue import. Sources: no-human:adapters.
Formats [supported] SQLite, YAML and git. Sources: no-human:config.
Cancellation [unsupported] No account to cancel. Sources: no-human:site.
Account independence [supported] Fully usable without a vendor account. Sources: no-human:site.
History portability [partial] History lives in the local database; PRs live on the git host. Sources: no-human:config.
Code and prompts go to the chosen model providers and your git host; state stays in local SQLite; credentials in a 0600 .env; telemetry is on by default through two channels: server-side usage events with a closed documented allowlist, and browser-side PostHog autocapture, dead clicks, heatmaps, performance, exceptions and masked session replay of the app's own interface (a masked-surface guarantee the docs say is not a claim that replay cannot capture content); the one opt-out is telemetry.enabled false in config.yaml, with no onboarding prompt or Settings toggle; official builds forward the onboarding email to a hosted intake; the coder can send anything its commands send.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
15 atomic answers
Code data [supported] Local worktrees; sent to the provider and git host. Sources: no-human:security.
Prompts [partial] Server-side events carry no prompts; browser session replay records the app's interface with operator-content blocks masked by hand-applied ph-no-capture, which the docs describe as a masked-surface guarantee, not proof it cannot capture content. Sources: no-human:telemetry, no-human:telem-web, no-human:telem-doc.
Credentials [supported] Stored in ~/.no_human/.env mode 0600, never in repo or config. Sources: no-human:security.
Subprocessors [partial] PostHog and the registration endpoint are named; the model provider is the user's. Sources: no-human:telemetry, no-human:email.
Telemetry disclosure [partial] TELEMETRY.md lists every server event and the browser screen_viewed; docs/configuration.md separately describes browser autocapture and session replay and states the nine-event table does not cover them. Sources: no-human:telemetry, no-human:telem-doc.
Telemetry defaults [supported] On by default (opt-out). Sources: no-human:config.
Telemetry controls [supported] telemetry.enabled false and POST /api/telemetry/consent. Sources: no-human:config, no-human:api.
Consent prompt [unsupported] No onboarding prompt or Settings toggle: config.py states the consent step and Usage insights pane were removed and config.yaml telemetry.enabled false is the one opt-out. Sources: no-human:telem-web.
Free (Local) edition: $0 forever with no card and unlimited tasks, billed only by the user's model provider; Pro (flat per seat) and Max (cloud) are waitlist-only with prices to be announced; Enterprise is by contact.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free Local; paid tiers waitlisted · 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
See atomic answers for the unmet or unknown parts.
12 atomic answers
Currency [unknown] No price is published, so no currency. Sources: no-human:site.
Interval [unknown] Pro is described as flat per-seat subscription; interval not stated. Sources: no-human:site.
Seats [partial] Pro is per seat; Local has no seats. Sources: no-human:site.
Seat minimum [unknown] Not published. Sources: no-human:site.
Workspace fees [unknown] Not published. Sources: no-human:site.
Included usage [unsupported] Local includes no model usage; billed by the provider. Sources: no-human:site.
Credits [unsupported] No credits. Sources: no-human:site.
Trial [unsupported] Local is free forever; paid tiers are waitlisted. Sources: no-human:site.
Overages [unknown] Not published. Sources: no-human:site.
Taxes [unknown] Not published. Sources: no-human:site.
Model cost [supported] Billed by the user's provider, not through no_human. Sources: no-human:site.
Compute cost [supported] User's own machine for Local; Max cloud compute pricing unpublished. Sources: no-human:site.
Platform cost is $0 for one, two or five developers on the Local edition (no collaboration features); model costs per task are estimated in-app but no scenario prices are published; Pro and Max prices are unannounced.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
11 atomic answers
One person [supported] $0 platform cost. Sources: no-human:site.
Two people [partial] $0 platform cost; each runs a separate local install without sharing. Sources: no-human:site.
Five people [partial] $0 platform cost; Pro team brain is unpriced. Sources: no-human:site.
Task assumptions [unknown] No scenarios published; none invented. Sources: no-human:site.
Generally available free local edition (v0.2.4 on 2026-09-17, six releases since 2026-09-07); pre-1.0; macOS, Linux and Windows builds with Windows and Linux items marked unverified; waitlist-only paid tiers; Discord and GitHub issues; SECURITY.md; no response commitments or status page.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Release activity [supported] Releases v0.2.1 to v0.2.4 between 2026-09-07 and 2026-09-17; default branch active to 2026-10-09. Sources: no-human:release, no-human:repoapi.
Status history [unknown] No status page. Sources: no-human:site.
Deprecation [partial] Keep a Changelog entries; no deprecation policy. Sources: no-human:changelog.
Export options [partial] API and PRs. Sources: no-human:api.
Free forever Local edition with no card; user pays model provider usage and own hardware; BYO Claude or ChatGPT subscriptions work; remote compute, phone and team collaboration are not in the free edition (paid tiers waitlisted).
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
Trial [unsupported] No separate trial. Sources: no-human:site.
Credits trial [unsupported] No credits. Sources: no-human:site.
Self host cost [supported] Local hardware and model provider spend. Sources: no-human:site.
Paid only [unsupported] Not paid-only. Sources: no-human:site.
Early access [partial] Pro and Max are gated by waitlist. Sources: no-human:site.
Free scenarios [partial] BYO subscriptions, issue intake and review work free; remote compute, mobile and team collaboration do not exist in Local. Sources: no-human:site, no-human:security.
Card required [unsupported] No card. Sources: no-human:site.
Usage caps [unsupported] Unlimited tasks, limited by model spend. Sources: no-human:site.
Paid evaluation [partial] A Claude or OpenAI credential is needed to evaluate. Sources: no-human:security.
Partially specified: Local is free with stated terms; Pro is flat per seat and Max cloud have no price yet; Enterprise is quote-only; model costs pass through unmarked.
Scope
Surfaces: Web board on 127.0.0.1:8420 · Version: v0.2.4 (commit a00adde, released 2026-09-17) · Plan: Free local edition (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
See atomic answers for the unmet or unknown parts.
9 atomic answers
Units [partial] Per seat for Pro; none for Local. Sources: no-human:site.
Included limits [partial] Local unlimited; Pro and Max limits unannounced. Sources: no-human:site.
Overages [unknown] Not published. Sources: no-human:site.
Compute markup [unknown] Max cloud compute pricing is unannounced. Sources: no-human:site.
Model markup [supported] Inference is billed by the user's provider, not through no_human. Sources: no-human:site.
Minimum commitment [unknown] Not published. Sources: no-human:site.
Cancellation [unknown] Not published. Sources: no-human:site.
Sales only unknowns [supported] Enterprise is by contact; Pro and Max are waitlist-only. Sources: no-human:site.
Transparency classification [supported] Partially specified: Local terms stated; paid tiers have named missing prices. Sources: no-human:site.
Free (Local) forever with no card, billed by the user's own model provider; Pro (team brain, per seat) and Max (cloud) are waitlist-only with prices to be announced; Enterprise adds SSO, SCIM, audit trail and budget policy by contact; desktop for macOS, Windows, Linux; no compliance certification claims.
Three coder backends: claude (Agent SDK to claude CLI), codex (codex exec), local (claude CLI at your own server); planner, supervisor and intake stay on Claude; reviewer defaults to Claude and can be set per role.
Subscription or API-key billing path scrubbing; agent never merges (approve is human); coder runs bypassPermissions by default with acceptEdits option; network egress depends on agent commands.
Kind
documented
Locator
1 One billing path; 2 The agent never merges; 7 What leaves your machine
Five-minute path: install prerequisites, nh init, nh task add from title or issue URL, run in foreground, review and approve, parallel overnight drain with --until-empty.
Kind
documented
Locator
Install, nh init, add first task, review and approve, overnight drain
Plan, implement, review loop bounded by max attempts; reviewer is a fresh-context second model; tamper guard counts deleted tests before review; reproduction gate; merge-ready policy; limits section lists what it does not do.
The pool width is clamped to a CPU-derived ceiling (cpu_count // 3, minimum 2) when the server starts, and each task's pytest-xdist auto worker count is bounded to cpu // max_workers; Settings reports cpu_count and hardware_ceiling. No memory limit or live load check was found in these modules.
The web board has a phone layout keyed to a 640px media query shared between JavaScript and styles.css (pinned by useIsPhone.test.mjs); the board API stays unauthenticated and loopback-checked.
Task rows and cards open the task drawer with Enter or Space, dialogs trap Tab focus and Escape closes slide-overs; the UI audit documents keyboard fixes. No list of keyboard shortcuts or command palette was found.
Kind
source-inspected
Locator
onKeyDown on task rows (Enter and Space); docs/UI_AUDIT.md B2 and M13 rows
No release, deploy, rollback or revert command, endpoint or gate found. Matches are classifier vocabulary lists, commit-guard and scheduler comments about transaction rollback and reverted status writes, and a 'no automatic rollback' note on nh doctor --fix-walks. No live app preview or evidence gallery exists: 'preview' matches are text result previews of tool calls in the activity log, and 'recording' matches are the optional product telemetry's masked session replay, not delivery evidence.
Kind
source-inspected
Locator
grep -rniE 'gh release|create_release|git tag|deploy|rollback|revert|gallery|preview|recording' over src/no_human (Python) and web/src, plus the nh command list in cli/commands.py
The desktop app checks GitHub Releases at startup once a day without downloading; neither NH_NO_UPDATE_CHECK nor updates.enabled reaches that path, so the only way to stop it is not to run the desktop app (documented as a defect).
Kind
documented
Locator
One GET to pypi.org per day; The desktop app checks GitHub Releases at startup
With telemetry.enabled true (the default) the browser board loads PostHog with autocapture, pageview, dead-click, heatmap, performance and exception capture and masked session replay of the app's own interface; inputs are masked and operator-content elements carry a hand-applied ph-no-capture block. config.py states the onboarding consent step and Settings pane were removed and config.yaml telemetry.enabled false is the one opt-out.
Kind
source-inspected
Locator
file header comment: ALL CHANNELS ON, MASKED REPLAY; config.py telemetry block and TELEMETRY_CONSENT_QUESTION
The nine-event table covers only what the app sends via posthog.capture. Autocapture, dead clicks, heatmaps, web vitals and exceptions are collected alongside, and session replay is a masked-surface guarantee, not a claim that replay cannot capture content; it can capture pixels of anything not on the masked list. docs/security.md adds that the desktop update check cannot be disabled by NH_NO_UPDATE_CHECK or updates.enabled.