clawnify_app_api_delete
ChatGPTSend a DELETE request to a deployed Clawnify app and return its response. It reaches the Clawnify app's own HTTP API: every app documents it at GET /llms.txt (a lean index of every endpoint) and GET /api/openapi.json (full request/response schemas); read those with clawnify_app_api_read first. Only an older app that serves neither documents itself in agent_instructions, via clawnify_get_app.
clawnify_app_api_read
ChatGPTSend a GET request to a deployed Clawnify app and return its response. It reaches the Clawnify app's own HTTP API: every app documents it at GET /llms.txt (a lean index of every endpoint) and GET /api/openapi.json (full request/response schemas); read those with clawnify_app_api_read first. Only an older app that serves neither documents itself in agent_instructions, via clawnify_get_app. List endpoints are paged: pass the app's own paging params (limit with offset or page) and its search/filter params rather than fetching everything. JSON/text responses are returned inline; binary responses are summarized (size + type), not inlined. To read one value of a large JSON response, such as a long text field of one record, pass field: a long string comes back in parts by offset and limit (characters), an array by items, each part with the next_offset to continue from.
clawnify_app_api_write
ChatGPTSend a POST, PUT or PATCH request to a deployed Clawnify app, which creates or changes its data, and return its response. To change part of one long text field of a record without sending all of it, pass field and edits instead of body: the record at path is read with GET, each edit replaces one exact old_text with new_text, and only field is written back to path with method (PUT or PATCH), so use a route that updates just the fields it is sent. Each old_text must occur exactly once; selections cannot overlap; all refer to the text as it is now, and the batch saves together; an empty new_text deletes. The read and the write are two requests, so a change made to that field in between is overwritten. Read the text first with clawnify_app_api_read and field. It reaches the Clawnify app's own HTTP API: every app documents it at GET /llms.txt (a lean index of every endpoint) and GET /api/openapi.json (full request/response schemas); read those with clawnify_app_api_read first. Only an older app that serves neither documents itself in agent_instructions, via clawnify_get_app.
clawnify_create_app
ChatGPTBuild a new app for this workspace from a description. Returns immediately with an app_id; the build runs in the background, so poll clawnify_get_app for status. Asks the person first.
clawnify_delete_app
ChatGPTDelete an app, its database and its stored source. This cannot be undone. Asks the person first.
clawnify_docs_search
ChatGPTSearch across the org's apps' procedure docs. Returns hits ranked by relevance, each with app slug, procedure name (if applicable), and a snippet.
clawnify_get_agent_instructions
ChatGPTRead an agent's AGENTS.md (its authored instructions) plus a content hash. Read this before editing, and pass the hash back to clawnify_set_agent_instructions as base_hash so your write won't clobber a version that changed since. Read-only. Pass agent_id only when the org has more than one agent.
clawnify_get_app
ChatGPTGet one app: build status and any build error, its URL, whether it has unpublished changes, the integrations it uses, template provenance, and its agent_instructions. Poll it after create or update.
clawnify_get_flow
ChatGPTRead one flow's full DRAFT definition, plus a content hash and published_version. Pass the hash back to clawnify_write_flow as base_hash so your edit can't clobber a version that changed since. The draft is what you edit; whether it is also what RUNS depends on published_version — a number means that published version runs instead, null means this draft is what runs. Read-only.
clawnify_get_flow_runs
ChatGPTInspect flow executions. Without run_id: a summary list of runs, newest first, 20 per page (limit, then cursor for the next page), optionally filtered by flow name and status (running | completed | paused | waiting | failed | cancelled). With run_id: that run in full (status, state, per-node trace, approval info — how to see where a flow failed). A run too large to return whole comes back as an outline of its fields with their sizes; read the part you need with path. Read-only.
clawnify_get_session_transcript
ChatGPTRead one session's transcript — a bounded tail of turns (normalised messages on OpenClaw 2026.9+ and Hermes, raw JSONL entries on older OpenClaw) — to see exactly where an agent veered, looped, or burned tokens. key is the agent-scoped session key from clawnify_list_agent_sessions (e.g. "agent:ash:lead:row_8268"). Read-only. Pass agent_id only when the org has more than one agent.
clawnify_get_usage
ChatGPTReturn current billing / quota info for the caller's org.
clawnify_list_agent_sessions
ChatGPTList an agent's stored OpenClaw sessions, including its sub-agents', newest first, with token counters — a high contextTokens flags a bloated, expensive, drift-prone session. Read-only; the first step in diagnosing a misbehaving agent. Pass agent_id only when the org has more than one agent — discover them with clawnify_list_agents.
clawnify_list_agent_skills
ChatGPTThe EFFECTIVE skill inventory an agent can actually run — workspace + shared managed-root + bundled ∩ allowlist, with eligibility — as OpenClaw's loader reports it, not just what sits in the agent's workspace folder. Use it to check whether an agent already has the skill a task needs before authoring or granting one. Read-only. Pass agent_id only when the org has more than one agent.
clawnify_list_agents
ChatGPTList the org's AI agents. Each entry is one agent (the "employee"), with a sub_agents array of the narrower helpers it delegates to — that is the whole hierarchy, nothing above or between — plus flows (the ClawFlow automations on its box: name, trigger, node count, published version) and integrations (the services it can use: per-agent connections plus org-shared ones, scope-tagged). Start here to see which agents exist and which agent_id to pass to the session / transcript / skill tools. An agent missing from this list does not exist in this org.
clawnify_list_apps
ChatGPTList apps the caller can access in the current org.
clawnify_list_flows
ChatGPTList the ClawFlow automations on an agent's box: name, description, trigger, declared inputs, node tree, and published — the version number the box runs. null means the flow has never been published, in which case the box runs its DRAFT, so editing that flow changes production immediately. Read-only; start here before reading or editing a flow.
clawnify_list_integrations
ChatGPTList the services this workspace has connected, so you know what you can actually reach before searching for actions.
clawnify_list_orgs
ChatGPTList the Clawnify organizations this connection can act on, with each one's name and handle. When there are several, every other tool takes an org argument naming one of them.
clawnify_publish_app
ChatGPTPublish an app's current draft so the workspace's people see it live. Asks the person first.
clawnify_set_agent_instructions
ChatGPTReplace an agent's AGENTS.md (its authored instructions) with new full text. Reversible (write again): the result returns the text before and after the change. Pass the hash from clawnify_get_agent_instructions as base_hash so a concurrent edit is rejected rather than overwritten. Takes effect on the agent's next session, no restart. AGENTS.md only — skills are edited via the clawnify CLI, flows via clawnify_write_flow. Pass agent_id only when the org has more than one agent.
clawnify_update_app
ChatGPTChange an existing app by describing the change. Queues a rebuild; poll clawnify_get_app for status. Asks the person first.
company_knowledge_archive
ChatGPTArchive a Company Knowledge document (owner/admin only): it disappears from search for every agent and tool but keeps its version history, and company_knowledge_restore brings it back. Use for docs that are superseded or no longer apply; for a doc that is merely wrong, prefer company_knowledge_review { state: "outdated" } so it can be fixed.
company_knowledge_get
ChatGPTRead one Company Knowledge document by id, including its full content — for when a search chunk isn't enough. Optionally fetch a specific published version (the version history is the audit trail for "what did the policy say when"). Cite as: Per <Title> v<N>.
company_knowledge_list
ChatGPTList Company Knowledge documents with structured filters (no semantic search). Useful for browsing the corpus: titles, types, status, versions, and freshness (review_state, published_at). Agent callers only see published documents; human callers also see drafts awaiting review. Knowledge hygiene is one call each: { review_state: "outdated" }, { review_state: "verification_expired" }, { review_state: "verification_requested" }, or { published_before: <ISO> } for docs nobody has re-published in a while.
company_knowledge_propose
ChatGPTPropose a Company Knowledge document as a DRAFT for human review — it is never published automatically. Use when something deserves to become official, citable company knowledge: a distilled policy interpretation, an FAQ-shaped answer, a procedure worth canonizing, or "remember this for the whole company" requests. For a targeted text edit, pass supersedes, expected_revision, and edits instead of content; unchanged title and metadata are inherited. To revise an existing doc, pass supersedes with its id: the revision becomes that doc's next version, so a title never needs a version number. A person with the owner or admin role can publish the draft in the conversation with company_knowledge_publish and the draft id; otherwise a reviewer publishes it from the dashboard review queue. Not for secrets or credentials; per-customer facts belong in memory, not Company Knowledge.
company_knowledge_publish
ChatGPTPublish a Company Knowledge draft (human callers with owner/admin role only). Bumps the version, snapshots it to the audit trail, and embeds it — the document becomes citable as <Title> v<N> by every agent and connected AI tool in the org, and searchable within seconds (indexed_at in the response is null while indexing finishes in the background). A pending revision (supersedes set) merges into its target document on publish; the draft row is removed.
company_knowledge_restore
ChatGPTRestore an archived Company Knowledge document (owner/admin only). It is re-published as the next version and becomes searchable again within seconds.
company_knowledge_review
ChatGPTRecord whether a PUBLISHED Company Knowledge document still holds (human callers with owner/admin role only). verified = a human vouches for it today — optionally time-boxed with until (e.g. 90 days out); when that passes the doc flips to verification_expired, is demoted in search, and its owner is emailed. outdated = the content is wrong or stale; agents stop seeing it in search until it is revised and re-verified — say why in note. verification_requested = unsure, ask the owner. none clears. Use when the human confirms a doc is current, notices one is wrong, or asks to audit the corpus.
company_knowledge_search
ChatGPTSemantic search over the organization's official, citable Company Knowledge corpus (published documents only — policies, SOPs, pricing, brand, FAQs, legal). It is the organization's source of truth for company facts. Every result carries a citation in the form <Title> v<N>, such as "Refund Policy v3". Each result also carries review_state and published_at: verified docs are ranked up; verification_expired docs are demoted and may be out of date; outdated docs never reach agent callers. Each result also carries its folder (the doc's home in the browse tree) — to list everything in a folder rather than rank by relevance, use company_knowledge_list { folder }.
company_knowledge_write
ChatGPTCreate or update a Company Knowledge document directly (human-authenticated callers only). For targeted text changes, pass id, expected_revision, and edits instead of content. Pass id to update an existing document's fields; omit it to create a new draft. A PUBLISHED document is never overwritten in place: a title/description/content edit is saved as a pending revision — the response carries revision_of and a NEW draft id — and you publish that draft id to merge it in as the next version. Saving alone never publishes; pass publish: true (owner/admin) to save and publish in one call once the human has confirmed, or call company_knowledge_publish later. Attempt this call when you need it; if the connection is not human-authenticated the server rejects it and names company_knowledge_propose as the fallback.
memory_delete
ChatGPTPermanently delete a memory from both Postgres and the vector index.
memory_get
ChatGPTRead one memory by id, including its full content.
memory_list
ChatGPTList memories with structured filters (no semantic search). Useful for browsing recent memories, listing pinned items, or filtering by type/scope.
memory_pin
ChatGPTPin or unpin a memory. Pinned memories appear in the pinned pack and are always injected at session boot. Use sparingly; the pack has a budget.
memory_pinned
ChatGPTFetch the agent's pinned memory pack — a small set of memories that are always injected at session boot. Useful for refreshing the pack mid-session.
memory_recall
ChatGPTSemantic search across the agent's accessible memories. Returns the most relevant memories (with scores) for a natural-language query. Use this before answering anything that depends on what was learned earlier.
memory_write
ChatGPTSave a durable memory the agent (or user) wants to recall later. Default scope = the caller's own private scope (the agent for agents, the user for users) — override via scope_type + scope_id. Include description (one-line summary, mandatory) and content (full body). See clawnify-protocol-memory.md for type semantics.
project_add_document
ChatGPTAdd a markdown document to a Project — scoped reference for that client/engagement (e.g. notes, a brief, a distilled summary). Unlike company_knowledge_propose, this is not governed/citable org knowledge; it's working context for the Project. Use [[wikilinks]] to relate documents.
project_delete_document
ChatGPTDelete a document from a Project for good: its versions and search index go, and for an uploaded file its original too. It cannot be undone. Only a person with the owner or admin role can; to correct a document, edit it with project_update_document instead. Use when the person asks to remove a document, or for one added by mistake.
project_list_documents
ChatGPTList the documents and files in a Project (no semantic search) — titles and descriptions, for browsing what a client/project contains.
project_search
ChatGPTSemantic search within a single Project (a specific client/engagement's documents and files) — use when the question is about that client/project, not the whole company. Get the project_id from projects_list. For org-wide facts (policies, pricing, brand), use company_knowledge_search instead.
project_update_document
ChatGPTEdit a document in a Project: change part of its body with exact-text edits, replace the whole body with content, or rename it with title. Saved and searchable at once, with no review step; the earlier body is kept as a version. Read the document first with company_knowledge_get { id }, which returns its body and edit_revision, and pass that as expected_revision: a document changed since you read it fails without writing. Get ids from project_list_documents.
projects_list
ChatGPTList the organization's Projects — scoped knowledge containers for specific clients, engagements, or capabilities (distinct from company_knowledge, the org-wide source of truth). Returns each Project's id, name, and kind. Use this to discover which Project to search or add to.
tasks_complete
ChatGPTMark a task on the Tasks board as completed (or reopen it with completed: false). Use tasks_list to find the id.
tasks_create
ChatGPTCreate a task on the organization's Tasks board. Assign it to a person (assignee_user_id, an org member — use clawnify_list_agents for agents and the org's members for people) or to an agent (assignee_server_id): an agent task is dispatched into that agent immediately and the agent does it. Leave both unset for an unassigned task. Returns the task, including run_status when it went to an agent.
tasks_delete
ChatGPTDelete a task from the Tasks board. Irreversible — only for a task created by mistake (a duplicate you just made) or when the user asks. To close finished work use tasks_complete instead.
tasks_list
ChatGPTList the organization's tasks from the Tasks board: every open task plus recently completed ones, with the people and agents rosters a task can be assigned to (ids and names). Each task has a title, an optional due_date (YYYY-MM-DD), one assignee — either assignee_user_id (a person) or assignee_server_id (an agent) — completed_at, and for agent tasks a run_status (queued | running | done | failed) with run_result, the agent's own summary of what it did. Use it to answer "what is on X's plate" or to find a task before completing it.
tasks_update
ChatGPTEdit a task on the Tasks board: title, due_date, or who it is assigned to. To unassign, pass assignee_user_id: null and assignee_server_id: null. Moving a task onto an agent dispatches it to that agent; taking it off an agent mid-run stands the agent down. Use tasks_list to find ids and the people / agents roster.