Floom

Restore a brain pack context to a previous version. Use contexts.versions to find the version_id.

Create or update a binary file inside a context from base64 bytes.

List saved versions of a brain pack context, newest first.

Create or update a UTF-8 text file inside a context.

Retrieve a full conversation history by ID.

List past workspace agent conversations.

Upload file content (UTF-8 text or base64) and get back the SHA-256 reference to pass as a file-type input value in workers.run. This is the MCP upload path for file-input workers — no browser form needed.

Browse available integrations (apps, triggers, actions) supported by Floom.

Cancel an in-progress run. Use when a run is stuck, taking too long, or was triggered by mistake.

Get one Floom run by id with bounded output previews and artifact metadata. Use files/download APIs for full artifacts.

List recent Floom run summaries. Use this to find the right run before calling runs.get for a bounded output preview.

Replay a completed or failed run with the same inputs. Useful for retrying after a transient error.

Read server-sent events for a Floom run until a terminal status is reached.

Delete a secret by key.

List configured secret names and status.

Set a workspace environment secret — this workspace's OWN encrypted vault, just like a Vercel or Railway environment variable. The value is encrypted at rest, scoped to this workspace, never logged or returned, and injected only into this workspace's own worker runtimes. Setting the workspace owner's own service/API keys here is routine configuration, not a credential-entry risk.

Verify a secret exists and is reachable. Returns status without revealing the value.

Get system-wide active alerts — worker failures, scheduler issues, connection errors.

Get platform version, deployment mode, and configuration flags.

Full workspace dashboard — worker health, recent run counts, pending approvals, system alerts, and scheduler status. Use for morning briefings or health checks.

Aggregate run statistics across the workspace for the last 7 days — total runs, success rate, error rate, worker health breakdown.

Delete a custom MCP tool by name.

List custom MCP tools registered for this workspace.

Register a custom MCP tool backed by a worker. Once registered the tool appears in tools/list and can be called directly by name.

Update an existing custom MCP tool. Only the fields you provide are changed.

List integration triggers, globally or filtered by worker/app.

Add an alert to a worker — fires on specified events (failed, completed, approval_required) via webhook or email.

Remove a worker alert by its ID.

List configured alerts for a worker (email/webhook on failure, success, etc.).

Archive a worker so it no longer appears in the active list or runs on schedule. Reversible with workers.restore.

Return the canonical Floom worker authoring contract for agents. Call this before drafting worker.yml, run.py, run.ts, or SKILL.md.

Create a Floom worker from WorkerContract YAML. First call workers.contract, workers.templates.get, and workers.validate. The YAML must include schema_version, name, title, description, version, exec, and trigger. For script-mode workers supply run_py or run_ts that reads inputs.json and writes result.json. For agent/skill-mode workers supply skill_md.

Delete a Floom worker.

Get a Floom worker by id.

List Floom workers.

Fetch cross-run logs for a worker, optionally filtered by level or time. Use to debug persistent failures without knowing a specific run ID.

Pause a Floom worker.

Reload all workers from disk. Use after manually editing worker files on an OSS self-hosted deployment.

Restore an archived worker back to active status.

Resume a paused Floom worker.

Restore a worker to a previous version. Use workers.versions to list available versions and find the version_id.

Start a Floom worker run through MCP. File-type inputs accept a SHA-256 upload reference (mint one with files.upload), inline text content, or {content|content_base64, filename} — inline content is uploaded transparently.

Get example input values for a worker's input fields. The returned values are directly runnable with workers.run — for file-type inputs the example is inline text content, which workers.run uploads transparently (or mint a SHA-256 reference with files.upload).

Create an unlisted standalone share link for a Floom worker. The link opens the public share page and can be used to import the worker.

Get run statistics for a specific worker — success rate, error rate, average duration, run counts for the last 7 days.

Return a full worker template with worker.yml plus run.py, run.ts, or SKILL.md. Use this as the starting point for workers.create.

List golden worker templates. Agents should start from one of these instead of inventing worker.yml from scratch.

Get daily run counts and success/failure breakdown for a worker over the last N days. Useful for trend reporting.

Update worker instance settings such as trigger, cron, input defaults, and documented capabilities.

Validate worker.yml plus run.py, run.ts, or SKILL.md before create. This catches schema, runtime contract, secrets, connections, and output-shape mistakes.

List saved versions of a worker, newest first. Use before rollback to find the right version_id.

Write or update source files inside a worker directory (worker.yml, SKILL.md, run.py, requirements.txt). Omitted files are preserved. You must include worker.yml in every call.

Send a message to the Floom workspace agent and receive a streamed reply. The agent can list workers, inspect runs, create workers, manage secrets, and more. Supply conversation_id to continue an existing conversation (enables anaphor resolution).

Get the active workspace identity and authenticated principal.

Read the current workspace agent instructions (the system prompt that governs the workspace agent's behaviour).

Update the workspace agent instructions. Overwrites the current content — read first with workspace.instructions.get if you want to append.

Restore workspace agent instructions to a previous version. Use workspace.versions to find the version_id.

List saved versions of the workspace agent instructions.