Reference

The command line, in help order.

The full inventory of the kastel command, generated from the engine itself. No line is written by hand, so no line can lie about a command that no longer exists.

A generated page

The summaries on this page are the ones kastel --help prints, published as they stand. You are reading exactly what your terminal will answer.

Where it runs

The kastel command works in the current directory: it reads your configuration, your governance rules and your content from there. Run it from your deployment folder, the one kastel init created.

kastel --help and kastel <command> --help print the full option list. This page is the inventory, in the same order.

The commands

Direct commands first, then the groups. The order is the help's, not a thematic ranking: the engine fixes it, and does so deliberately.

Commands

`kastel check`	Print budget status and a light resolvable check of agents and skills.
`kastel initTARGET--root--client--departments--language--template--image--embeddings--behind-proxy--force`	Create a ready-to-use client deployment (Brain skeleton + config + governance).
`kastel askQUESTION--task--consumer--person--k--show-chunks`	Ask the Brain a question and get a cited answer (read-only consultation).
`kastel export--output`	Archive the deployment data dirs into a portable, verifiable tarball.
`kastel backup--output--verify`	Write the provable continuity backup of this deployment.
`kastel restoreTARBALL--dry-run`	Restore a `kastel export` tarball into THIS fresh deployment.
`kastel importFOLDER--dept`	Import a folder of text documents into the Brain (governed ingest each).
`kastel import-mcpCOMMAND--args--dept`	Import resources from an MCP server into the Brain (governed ingest each).
`kastel distill--limit`	Distill conversational threads into normalized, searchable notes (R1).
`kastel graph-distill--limit`	Propose LLM-deduced typed relations through the moderation queue.
`kastel serve--no-browser --no-no-browser`	Open the editable company map in the browser (sovereign local server).
`kastel mcp--consumer--transport--host--port--tls-cert--tls-key`	Expose the structured context to external AI runtimes over MCP (read-only).
`kastel upgrade--no-reindex`	Migrate this deployment's Brain structure to the latest version (G-7).
`kastel reindex--incremental`	Rebuild the retrieval index from the Brain markdown (full, idempotent).
`kastel describe--dry-run`	Replace fallback page descriptions with one LLM-redacted sentence each.
`kastel emit-usage`	Push unsent sealed usage events to the configured cloud console (one cycle).
`kastel connect--runtime--all--token--consumer--agent--departments--allow--deny--expires-days--app-url--lang--write-skills`	Emit the connection snippet + the CLAUDE.md/AGENTS.md instruction block.
`kastel lint--strict`	Lint the Brain: unresolved wikilinks (with their source page), stale compiled indexes, orphan pages, missing descriptions.
`kastel suggest-departments--model`	Detect and store the departments to suggest activating (needs a provider).
`kastel doctor--remediate--yes`	Inspect Brain hygiene and (optionally) run the governed remediation plan.

Command groups

`kastel ingest`Governed ingest channel: build the Brain from a file or an email export.

`kastel ingest fileSOURCE--dept`	Construct context from a single source file into the Brain (governed channel).
`kastel ingest emailPATH--mailbox-owner--limit`	One-shot import of an email export (mbox) into the Brain (M-1 / D39).
`kastel ingest failures--source--all--limit`	List what the last syncs and drops could NOT read (and why).
`kastel ingest resolveSOURCERESOURCE_ID`	Close one failure row by hand (source-deleted item, accepted gap).

`kastel token`Per-consumer scoped MCP tokens: issue, list, revoke (HTTP transport auth).

`kastel token issueCONSUMER_NAME--agent--departments--allow--deny--expires-days--quiet -q`	Issue a scoped token. The plaintext is shown ONCE and never stored.
`kastel token issue-humanEMAIL--expires-days`	Issue a person-scoped token for a human (M-6, spec section 9).
`kastel token issue-adminLABEL--scope--expires-days--quiet -q`	Issue an ADMIN token for an admin surface (spec-token-control-api 3.2, spec-score-api).
`kastel token list`	List tokens: id, consumer, shape, created, expires, revoked, last used, sessions.
`kastel token revokeID_OR_CONSUMER`	Revoke a token: effective on its very next request (no verdict cache).

`kastel identity`Human identity mapping: link a login email to an employee slug (free tier).

`kastel identity linkEMAILSLUG`	Link a login email to an employee slug (governance/identities.yaml).
`kastel identity unlinkEMAIL`	Remove a login email link (the offboarding/DPO lever).
`kastel identity list`	List the login email -> employee slug mapping (governance/identities.yaml).

`kastel editor`The Brain editor role: grant, revoke, list, and the per-department approval requirement (governance/editors.yaml, edit_approvals.yaml).

`kastel editor grantSLUG--department`	Grant the editor role (refused for a slug absent from governance/aor.yaml).
`kastel editor revokeSLUG`	Revoke the editor role (effective on the next request; the role is read live).
`kastel editor list`	List editor grants and the approval-required departments.
`kastel editor require-approvalDEPARTMENT--off`	Toggle the per-department approval requirement (regulated mode, default empty).

`kastel siem`SIEM webhook: stream the sealed audit chain to your SOC (opt-in).

kastel siem push Deliver undelivered sealed audit events to the configured SIEM (one cycle).

`kastel db`Database operations: apply migrations, import the legacy SQLite store.

`kastel db upgrade`	Apply pending migrations (alembic upgrade head) to KASTEL_DB_URL.
`kastel db import-sqlite--path--force`	One-shot import of the legacy SQLite store (usage -> ledger, audit -> audit).

`kastel credentials`Persisted runtime secrets, encrypted at rest in Postgres (ADR-009). At-rest protection only: whoever controls the process and the KEK can decrypt.

`kastel credentials putSCOPECONNECTORPRINCIPAL--attr`	Encrypt and store a secret (value read from stdin, never from argv).
`kastel credentials getSCOPECONNECTORPRINCIPAL--reveal`	Show a credential's metadata; --reveal prints the decrypted value.
`kastel credentials rotateSCOPECONNECTORPRINCIPAL`	Replace a secret's value (read from stdin); bumps rotated_at.
`kastel credentials rmSCOPECONNECTORPRINCIPAL`	Hard-delete a credential (offboarding, RGPD). No-op if absent.
`kastel credentials lsSCOPE--connector`	List stored credentials' NON-SECRET metadata (never values).
`kastel credentials rekey`	Re-encrypt every stored secret under a new KEK (KASTEL_CREDENTIALS_KEK_NEW).

`kastel jobs`The durable-sync job queue: work it (run), inspect it (ls), retry a failed or dead-lettered job (retry). Postgres-backed, no broker (ADR-011).

`kastel jobs ls--status--limit`	List jobs newest-first; surfaces dead_letter loudly (invariant 4).
`kastel jobs run--once--max-jobs--loop--interval--sync-interval-minutes`	Work the queue: one-shot drain (default/cron) or durable worker (--loop).
`kastel jobs retryJOB_ID`	Re-queue a failed or dead-lettered job for an immediate attempt.

`kastel aor`Edit the AOR (who decides/validates what) and agent memory access, validated (B3).

`kastel aor show`	Print the current governance model in plain language (never raw YAML).
`kastel aor previewOPARGS`	Show the plain-language BEFORE/AFTER of an operation, without applying it.
`kastel aor dry-runOPARGS--limit--all`	Report which pages an operation would hide or expose, for every principal.
`kastel aor applyOPARGS--actor`	Validate and govern-write an operation (owner-attributed, audited, reversible).
`kastel aor revertCOMMIT--actor`	Governed undo of a prior AOR edit commit (attributed, audited, re-validated).

`kastel moderation`The human moderation feed: review the Guard's flagged writes (list/show/approve/reject).

`kastel moderation list--status`	List moderation items: id, age, source, routed_to, rules, pages.
`kastel moderation showITEM_ID`	Show the full effect of an item: the diff or the payload, rules, provenance.
`kastel moderation approveITEM_ID--note--decided-by`	Approve an item: apply its effect all-or-nothing (git merge for a brain_diff).
`kastel moderation rejectITEM_ID--note--decided-by`	Reject an item: discard the effect, restore the challenged page to active.

`kastel lessons`The agent-memory loop: compile execution traces into governed lesson candidates (compile), prune old compiled traces (prune).

`kastel lessons compile--agent--since`	Distil uncompiled agent traces into lesson candidates through the Guard.
`kastel lessons prune--before`	Prune old COMPILED traces (operator-only, logged, never automatic).

`kastel digest`The weekly freshness digest: generate every question the lint + staleness scan raises (generate), notify each recipient with a count-only pointer to the platform (send), and manage per-person notification preferences (prefs). Keeps the Brain alive.

`kastel digest generate--week--dry-run`	Generate the week's impact-ranked questions and the value report.
`kastel digest send--week--force`	Notify each recipient with pending questions: a count-only pointer.
`kastel digest prefs--for--set-channel--email--slack-webhook-url`	Show or set one person's notification preferences (INV-27).
`kastel digest answerQUESTION_IDANSWER_TEXT--as`	Record a person's answer to a generated question as a governed write.

`kastel page`Page lifecycle operations (operator-only): archive a page.

kastel page archiveREL_PATH Archive a page (active -> archived). Operator action, never automatic.

`kastel erase`Right to erasure (RGPD): erase a stored raw copy (--raw) or a whole mailbox slug (--mailbox), flagging (never deleting) governed facts that cite it. `erase list` shows erasable mailbox slugs.

`kastel erase list`	List the mailbox slugs that have raw copies (erasable with --mailbox).
`kastel erase pending`	List source deletions flagged for admin review (retained, not yet erased).
`kastel erase approve--connector--path--requester`	Approve pending flags: erase the flagged raw copies via the sealed admin lever.
`kastel erase dismiss--connector--path`	Dismiss pending flags: KEEP the flagged content, just clear the flag (no erasure).

`kastel score`The Context Score: context recall@k + structure + optional answer fidelity, always with breakdown and N. Manage the question set with `kastel score questions`.

`kastel score questions`	The per-deployment question set the score is measured on (spec 3.2).
`kastel score questions addQUESTION--expected--department`	Add one active question to the deployment's question set.
`kastel score questions list--status`	List the question set (id, status, scope, expected pages, text).
`kastel score questions importFILE`	Import a question-set file (the G-10 path: a golden set becomes the demo set).
`kastel score questions confirmQUESTION_ID--question--expected`	Owner confirm/edit of a prefill candidate: pending -> active (spec 3.2).

`kastel eval`The owned eval set (memory/evals/): run the Context Score over your git-owned evals, list them, or lint them. `run` is free core.

`kastel eval run--k--snapshot--json`	Compile memory/evals/ -> question set, then compute the Context Score.
`kastel eval matrix--k--json`	Sweep the owned evals across the declared inference models (Kastel Enterprise).
`kastel eval sentinel--k--force--json`	Run the model-liquidity regression sentinel (Kastel Enterprise).
`kastel eval list`	List the owned eval set (id/scope/sensitivity/expected), from the markdown.
`kastel eval lint`	Structurally validate the owned eval set (spec 3.3). Non-zero on problems.

`kastel cost`Provider cost visibility (informative): estimate an onboarding, show ledger actuals.

`kastel cost estimate--corpus--emails`	Order-of-magnitude API cost of onboarding a corpus, BEFORE spending money.
`kastel cost show--month`	Show this month's ledger usage grouped by consumer (read-only actuals).

`kastel audit`The sealed local audit chain: replay a request trace.

kastel audit replayTRACE_ID Replay one request as its five audit points (read-only, local).

`kastel providers`The configured inference/embedding providers and their declared posture.

kastel providers posture Show the declared data-handling posture of the configured providers.

`kastel cockpit`AI-system cockpit: document and export how your AI system works (portable).

`kastel cockpit show`	Print 'how your AI system works' (skills, agents, runtime, organization).
`kastel cockpit export--output`	Export the portable AI-system bundle (ai-system.md + manifest.yaml).

`kastel onboard`Onboarding interview: build the company's structured Brain.

`kastel onboard start`	Start (or resume) the scoping interview and report Temps A readiness.
`kastel onboard prefill`	Derive known facts from the imported documents (so they are not asked again).
`kastel onboard ask`	Print the next interview question (None when the phase is covered).
`kastel onboard answerTEXT`	Answer the current question (the one `kastel onboard ask` shows).
`kastel onboard map`	Propose the org map (a draft to review before confirming).
`kastel onboard confirm--by--edits`	Validate the proposed map: write the General wiki + governance/aor.yaml.
`kastel onboard deptNAME`	Deepen one department (Temps C): its processes, tools, and wiki.
`kastel onboard status`	Show completeness, time-to-context, and what to do next.

`kastel template`Brain-templates: package a vertical's STRUCTURE from a deployment (package), review it element-by-element (review/finalise), and scan it against the source corpus (scan). The template travels, client data never.

`kastel template package--out--from--vertical--name--author--version`	Extract a template DRAFT from a deployment: structure only, content purged.
`kastel template review--template--approve`	List review units, or mark one reviewed (tier-2, INV-3).
`kastel template finalise--template--against`	Finalise a template - refused unless every unit is reviewed AND the anti-leak scan against the RECORDED source is clean (INV-3 + INV-4).
`kastel template scan--template--against`	BLOCKING anti-leak scan (INV-4): fail on any source element in the package.

`kastel entity`Entity fiche maintenance (list types, merge duplicates).

`kastel entity types`	List this deployment's entity types: name, origin, pages.
`kastel entity mergeSRC_SLUGDST_SLUG--type--src-type--dst-type--apply`	Merge a duplicate entity fiche into the canonical one.
`kastel entity numbered-folds`	List past auto-merges the ENG-145 numbering guard now refuses.

`kastel packs`Context packs: the task recipes under memory/packs/.

`kastel packs list`	List every pack with its validity status (INVALID packs show the reason).
`kastel packs checkSLUG`	Validate one or every pack; exit non-zero on any invalid pack (CI-friendly).

`kastel feedback`Read the human feedback left on Brain answers (up/down + comments).

kastel feedback list--rating--days--limit--json List the answers people rated, newest first (badly-rated ones by default).

`kastel read-audit`Who read what: aggregated access counters and per-person consultation.

`kastel read-audit entities--entity--consumer--days--json`	Aggregated read/write day-buckets: open accumulator + sealed events.
`kastel read-audit personSLUG--days--json`	One person's local consultation record: page opens + answer citations.

`kastel claims`Attributed claims: who asserts what, with what confidence.

`kastel claims extract--limit--quote--enqueue`	Extract attributed claims from the shared Brain pages (BYOK, cached).
`kastel claims list--holder--kind--page--min-weight--limit--consumer`	List extracted claims (deployer surface, AOR-filtered).
`kastel claims searchTEXT--kind--limit--consumer`	Search claim text (deployer surface, AOR-filtered).
`kastel claims review--limit`	List promotion candidates: durable, well-attributed claims.
`kastel claims promoteCLAIM_ID--to--by`	Propose ONE claim as a governed page edit (moderation feed decides).

`kastel contradictions`Contradiction probe: run history, open count, manual runs.

`kastel contradictions report--limit--json`	Run history: headline rate + CI, cache hit rate, items, cost.
`kastel contradictions scan--json`	Run the deterministic scan now (zero-LLM, free, no provider needed).
`kastel contradictions run--enqueue--quote`	Run the probe now (BYOK judge; respects the monthly budget cap).

`kastel automations`The six BYOK automatisms: what runs by itself, and switch it.

`kastel automations list--json`	What runs by itself on this deployment, and when a change would land.
`kastel automations onAUTOMATION_ID`	Switch one automatism ON. It spends the deployment's BYOK from then on.
`kastel automations offAUTOMATION_ID`	Switch one automatism OFF.

The connectors

The connector group follows a single template, which saves repeating the same list for each one. The exceptions below are measured, not declared.

`kastel connector`Continuous connectors (M365): connect a mailbox, run a sync, garbage-collect, or clean.

33 connectors. 31 of them take the same clean, connect, gc, offboard, sync.

asanaboxcal_comcalendlyclickup_docsconfluencecrispdocusigndropboxdropbox_signfirefliesfolkfrontgdrivegmailgonggrainguruhubspotjiram365mailchimpmondaynotionpandadocquickbookssageteamstrelloxerozoho_crm

Exceptions

obsidian without: offboard
slack plus: opt-in-privateopt-out-private

The launcher

The npm launcher installs and drives a self-hosted Kastel on your machine. It has its own handful of commands, and passes everything else through to the engine: kastel <engine command> runs exactly the commands in the table above, inside the container.

usage

npx @kastel.ai/cli - install and run your own Kastel (free, self-hosted, no account)

Install, the first time:
  npx @kastel.ai/cli [TARGET_DIR] [options]
                          install (or resume/start) a deployment

Then install the command itself, once - it is named 'kastel':
  npm install -g @kastel.ai/cli

Everyday, from inside a deployment directory:
  kastel status           deployment state and endpoints
  kastel start | stop     start / stop the running stack
  kastel upgrade          move to the current public release (explicit,
                          never automatic)
  kastel uninstall [--purge-data] [--force]
                          preview by default; --force applies
  kastel doctor [--server --domain <fqdn>] [--json]
                          check this machine, write nothing
  kastel <engine command...>
                          run any engine command in the container
                          (check, import, onboard, token, export, ...)

Install options:
  --server             server shape: public TLS via Caddy (requires --domain)
  --domain <fqdn>      the domain your Kastel will answer on (server shape)
  --lang en|fr         language of the deployment (default en)
  --client <name>      company name (default: the directory name)
  --departments a,b    departments to activate at scaffold time
  --local-embeddings   fully local retrieval embeddings (ollama; server shape)
  --port <n>           MCP port on 127.0.0.1 (default 8787, auto-remapped if taken)
  --no-connect         do not wire the AI clients on this machine
  -h, --help           this help

Exit codes: 0 ok, 1 unexpected, 2 usage, 3 precondition, 4 image pull,
5 scaffold, 6 startup/health, 7 AI-client wiring (the install itself succeeded).

The engine image is proprietary (Kastel Engine EULA) and runs on YOUR machine;
this installer is Apache-2.0. No account, no telemetry, no calls home - ever.

280 commands, generated from the engine command line, refreshed on 2026-08-23 (engine d1e57f93ef0e).

A question about integrating?

Write to us: a question asked by email gets an answer the same day. Our network of partner integrators can also help.

The command line, in help order.

Where it runs

The commands

Commands

Command groups

kastel ingestGoverned ingest channel: build the Brain from a file or an email export.

kastel tokenPer-consumer scoped MCP tokens: issue, list, revoke (HTTP transport auth).

kastel identityHuman identity mapping: link a login email to an employee slug (free tier).

kastel editorThe Brain editor role: grant, revoke, list, and the per-department approval requirement (governance/editors.yaml, edit_approvals.yaml).

kastel siemSIEM webhook: stream the sealed audit chain to your SOC (opt-in).

kastel dbDatabase operations: apply migrations, import the legacy SQLite store.

kastel credentialsPersisted runtime secrets, encrypted at rest in Postgres (ADR-009). At-rest protection only: whoever controls the process and the KEK can decrypt.

kastel jobsThe durable-sync job queue: work it (run), inspect it (ls), retry a failed or dead-lettered job (retry). Postgres-backed, no broker (ADR-011).

kastel aorEdit the AOR (who decides/validates what) and agent memory access, validated (B3).

kastel moderationThe human moderation feed: review the Guard's flagged writes (list/show/approve/reject).

kastel lessonsThe agent-memory loop: compile execution traces into governed lesson candidates (compile), prune old compiled traces (prune).

kastel digestThe weekly freshness digest: generate every question the lint + staleness scan raises (generate), notify each recipient with a count-only pointer to the platform (send), and manage per-person notification preferences (prefs). Keeps the Brain alive.

kastel pagePage lifecycle operations (operator-only): archive a page.

kastel eraseRight to erasure (RGPD): erase a stored raw copy (--raw) or a whole mailbox slug (--mailbox), flagging (never deleting) governed facts that cite it. erase list shows erasable mailbox slugs.

kastel scoreThe Context Score: context recall@k + structure + optional answer fidelity, always with breakdown and N. Manage the question set with kastel score questions.

kastel evalThe owned eval set (memory/evals/): run the Context Score over your git-owned evals, list them, or lint them. run is free core.

kastel costProvider cost visibility (informative): estimate an onboarding, show ledger actuals.

kastel auditThe sealed local audit chain: replay a request trace.

kastel providersThe configured inference/embedding providers and their declared posture.

kastel cockpitAI-system cockpit: document and export how your AI system works (portable).

kastel onboardOnboarding interview: build the company's structured Brain.

kastel templateBrain-templates: package a vertical's STRUCTURE from a deployment (package), review it element-by-element (review/finalise), and scan it against the source corpus (scan). The template travels, client data never.

kastel entityEntity fiche maintenance (list types, merge duplicates).

kastel packsContext packs: the task recipes under memory/packs/.

kastel feedbackRead the human feedback left on Brain answers (up/down + comments).

kastel read-auditWho read what: aggregated access counters and per-person consultation.

kastel claimsAttributed claims: who asserts what, with what confidence.

kastel contradictionsContradiction probe: run history, open count, manual runs.

kastel automationsThe six BYOK automatisms: what runs by itself, and switch it.