tools

Creates a new agent with the specified configuration.

Deletes an agent owned by the current user.

Compares two version snapshots of an agent and returns a structured field-level diff covering model, system prompt, tools, sub-agents, and evals.

Returns full configuration for a specific agent including tools and sub-agents.

Lists all agents owned by the current user.

Returns the available model groups and model IDs for agent configuration.

Returns the available tool registry with IDs, names, descriptions, and groups.

Lists agent configuration version snapshots (newest first) with model, prompt, tools, sub-agents, and evals captured at each point.

Reverts an agent's configuration to a previous version snapshot. Creates a new version reflecting the reverted state.

Updates an existing agent's configuration.

Replaces an agent's starter suggestions shown in empty chats.

Returns the current date and time in the user's timezone.

Summarizes an agent's chat spend: total cost, average cost per turn, a per-model breakdown, and the costliest conversations, with an optional date floor.

Creates a new eval for an agent and snapshots a new version reflecting the change.

Deletes an eval and snapshots a new agent version reflecting the removal.

Lists historical eval runs for an agent, newest first and paginated. Optionally filters to a single eval. Each run carries its score, output, rationale, the agent version it ran against, and the conversation id of its trace.

Lists evals for an agent with their latest run summaries (score, output, rationale, per-trial aggregate).

Runs every eval for an agent in one batch (up to 3 at a time) and records each run with its score.

Runs a single eval against the agent's current configuration, records the run with its score, and saves its execution as a traceable conversation.

Updates one or more fields on an existing eval and snapshots a new version reflecting the change.

Reads files from public GitHub repositories in batch.

Delete a saved memory by id. Use when the user asks to forget a fact, or to clean up an outdated entry.

Save a durable fact about the user (preferences, identity, context) that should survive across conversations. Use for stable facts, not turn-by-turn details. Saved memories become searchable on the next turn after embedding.

Search saved user memories by semantic similarity. The chat already injects the top matches for the current user message; call this for targeted lookups that the auto-injection didn't surface.

Discovers movies on TMDB by genre, year, language, and sort order.

Discovers TV series on TMDB by genre, first-air year, language, and sort order.

Fetches full TMDB metadata for a movie by id.

Searches TMDB across movies, TV, and people in a single request.

Lists what's trending across movies, TV, and people on TMDB.

Lists trending movies on TMDB.

Lists trending TV series on TMDB.

Fetches full TMDB metadata for a TV series by id.

Returns the projected execution trace for a conversation: timed steps with tool calls, inputs/outputs, errors, and token usage.

Lists recent conversations for an agent with aggregated timing, event count, and cost. Returns summaries only; use traces-get for full per-step detail.

Fetches the contents of a URL and returns it as markdown, text, or HTML.

Searches the web (via Tavily) and returns a list of titles, URLs, and snippets.

Fetches a Wikidata item's labels, descriptions, statements, and sitelinks by Q-id.

Resolves Wikidata Q-ids and P-ids to labels and descriptions in batch.

Searches Wikidata for entities by label and aliases.