CLI reference
Every anymo verb, its synopsis, key flags, and one realistic example. Run anymo --help for the same surface inline. From a checkout, invoke the CLI as cargo run -p anymo-cli -- <verb>; this page uses the short installed form.
Conventions
--workspace <dir>selects the workspace (default./anymo-demoforsetupanddoctor,.otherwise). State lives in<workspace>/.anymo/.--json(where listed) emits a machine-readable report instead of the human view.- Permission modes are exactly
auto,manual,plan. There is no bypass. - Client verbs that talk to a running app-server take
--websocket <addr>or--unix-socket <path>plus--token <token>(orANYMO_APP_SERVER_TOKEN).
First run
setup
Verify the environment, scaffold a demo workspace with a sanitized config, sample skill, and first-run scenario, run an offline mock demo, and write a desktop-readable setup-state.json.
Flags: --workspace <dir>, --repo-root <dir>, --force (overwrite existing demo files), --json, --probe (probe model endpoints; off by default).
anymo setupdoctor
Check the environment and report exactly what to fix as a grouped OK/WARN/FAIL/SKIP report. Offline-first: a missing provider or unbuilt UI is a warning, not a failure.
Flags: --workspace <dir>, --repo-root <dir>, --json, --strict (warnings fail the exit code), --no-probe (skip network probes).
anymo doctor --strict --no-probeRuns
run
Execute a headless agent run against the mock provider or an OpenAI-compatible HTTP endpoint. With an endpoint flag (--websocket or --unix-socket) the run targets a connected app-server instead of running locally.
Flags: --provider mock|http, --mock (alias for --provider mock), --swarm, --workspace <dir>, --goal "…", --model <model>, --permission auto|manual|plan, --reasoning low|medium|high|xhigh|max, --fallback <base_url>,<model> (repeatable; http only). Defaults: --permission auto, --reasoning xhigh.
Provider environment for --provider http: ANYMO_PROVIDER_BASE_URL, ANYMO_PROVIDER_MODEL, ANYMO_PROVIDER_API_KEY; fallbacks via ANYMO_PROVIDER_FALLBACK_BASE_URL/_MODEL/_API_KEY.
anymo run --mock --workspace anymo-demo --goal "summarize my notes"runs
List recorded runs from the workspace event log: id, status, event count, goal.
anymo runs --workspace anymo-demoshow
Show one run's status, summary, and event timeline, plus any approvals still awaiting a decision with their approval ids. --json emits the same view as an object.
anymo show <run-id> --workspace anymo-demosessions (alias history)
Summarize the run-history event log: run count and the most recent sessions with terminal status. Content-free and support-safe.
anymo sessions --workspace anymo-demo --jsonGoals
goal
Set or inspect a session goal, a completion condition judged after every turn by a separate no-tools evaluator. goal "<condition>" sets it; bare goal shows status; goal pause|resume; goal clear (aliases stop|off|reset|none|cancel).
Flags: --workspace <dir>, --provider mock|http, --permission auto|manual|plan, --reasoning low|medium|high|xhigh|max, --max-turns <n>.
anymo goal "the report exists and cites three sources" --provider mock --max-turns 2 \
--workspace anymo-demoSkills
skill list
List the workspace's saved skills: id, version, name, tools.
anymo skill list --workspace anymo-demoskill run
Run a saved skill as a recorded run, binding parameters per step. A bare k=v binds to every step whose tool accepts k; a 1-based STEP.k=v binds to one step. A skill step can declare its own inputs, including {{steps.N.output}} templates that pipe an earlier step's output into a later one. Missing required inputs are all reported in one error.
Flags: <id> (positional), --param [STEP.]k=v (repeatable), --workspace <dir>, --permission auto|manual|plan.
anymo skill run page-report --param url=https://example.com --workspace anymo-demoApp-server and clients
app-server
Host the kernel over JSON-RPC so a desktop or remote front end can drive runs.
Flags: --workspace <dir>, --stdio | --unix-socket <path> | --websocket <addr>, --token <token>, --remote-runner (also drive the automation scheduler).
anymo app-server --workspace anymo-demo --websocket 127.0.0.1:9300 --token devtokenconnect (alias health-check)
Check a running app-server's health. Client flags: --websocket/--unix-socket, --token.
anymo connect --websocket 127.0.0.1:9300 --token devtokenlist-runs
List runs via a connected app-server.
anymo list-runs --websocket 127.0.0.1:9300 --token devtokenrun via app-server
When run carries --websocket or --unix-socket, it starts the run on the server and prints the run id plus a next-step hint: a manual run points you at anymo approvals, otherwise at tail-events.
anymo run --websocket 127.0.0.1:9300 --token devtoken --goal "summarize my notes" \
--permission manualapprovals
List approvals awaiting a decision on a connected app-server, printing each approval id (the value approve needs).
anymo approvals --websocket 127.0.0.1:9300 --token devtokenapprove
Respond to a pending approval. The default decision is approve.
Flags: --approval-id <id>, --decision approve|deny, --reason "…", plus client flags.
anymo approve --websocket 127.0.0.1:9300 --token devtoken --approval-id <id>tail-events
Stream a run's events via a connected app-server. Exits once the run reaches a terminal state, so tailing an already-finished run prints its outcome and returns instead of hanging.
Flags: --run-id <id>, --from-sequence <n>, plus client flags.
anymo tail-events --websocket 127.0.0.1:9300 --token devtoken --run-id <id>Artifacts
artifact
Artifact studio. demo emits a validated multi-artifact bundle locally; list and read query a connected app-server.
Flags: demo [--workspace <dir>] | list [--run-id ID] | read --artifact-id ID (list and read need client flags).
anymo artifact demo --workspace anymo-demoQuery Fabric
Local, deterministic retrieval over the workspace, memory, skills, and tools.
index
Build the local Query Fabric index. Flags: --workspace <dir>, --json.
anymo index --workspace anymo-demoquery
Search the index. Flags: QUERY (positional), --source SCOPE, --max-results N, --mode fast|balanced|deep, --json.
anymo query "provider failover" --workspace anymo-demotool-search
Rank tool specs for a query. Flags: QUERY, --max-results N, --json.
anymo tool-search "write a file"context-pack
Build bounded context packets for a query. Flags: QUERY, --json.
anymo context-pack "summarize the demo" --workspace anymo-demoDiagnostics
context (alias ctx)
Content-free report on the project context file (AGENTS.md): present, size, truncation. Never prints file content. Flags: --workspace <dir>, --json.
anymo context --workspace anymo-demomodels (alias model)
Explain or select the provider route. Flags: explain|select <route>, --candidate provider:model[:priority], --strategy priority|declaration-order, --unavailable P, --require CAP, --json.
anymo models explain default --jsonversion
Report binary, build hash, and protocol versions. Flag: --json.
anymo version --jsonDesktop control
computer
Local desktop control: observe, list-windows, focus-window, click, type, hotkey, clipboard-read, clipboard-write. The real OS backend requires building with --features computer-os; otherwise a mock backend runs. Flags: --approve, --plan, --json.
anymo computer observe --jsonProject harness
A set of verbs for running jobs in isolated Git worktrees. All take --repo PATH and --json.
| Command | Purpose |
|---|---|
anymo project doctor | Report harness readiness |
anymo worktree create JOB [--branch BRANCH] [--base REF] | Create a worktree-backed job |
anymo worktree list | List worktree-backed jobs |
anymo worktree clean JOB|--all [--remove-branch] | Remove a job's worktree |
anymo job run JOB [--command NAME] [--approve] | Run a job in its worktree |
anymo job status JOB | Show one worktree job |
anymo patch review JOB | Review a job's patch set |
anymo test run JOB [--command NAME] [--approve] | Run a job's test command |
anymo worktree create my-job --repo . --jsonRuntime Bridge
Coordinate external agent runtimes, or mount Anymo's brain into any MCP-capable agent. See Runtime Bridge for the full security model and configuration reference.
| Command | Purpose |
|---|---|
anymo runtimes [--json] [--project PATH] | List detected runtimes: id, version, surfaces, source, tier |
anymo runtime run NAME --goal "..." --project PATH | Drive a runtime end to end (the mock runtime works out of the box) |
anymo mcp-serve --workspace DIR [--allow-tools a,b] | Host the brain as a stdio MCP server (read-only by default; propose-not-write memory) |
# The built-in mock runtime needs no external dependency:
anymo runtime run mock --goal "demo the bridge" --project .
# Mount the brain into any MCP-capable agent:
anymo mcp-serve --workspace .