List available Vending Machine agents and their capabilities.

Hire a Vending Machine agent to perform a task.

Args: agent_id: Which agent to hire. One of: bug-hunter, test-goblin, devops-dwarf, cloud-sensei, code-gremlin, data-sprite, desk-pilot, embeddings-agent, inbox-zero, mcp-maker, number-crunch, pdf-forge, vibe-writer. task: Detailed task description. Be specific — the agent only sees this text.

Check the status of a background agent job.

Args: job_id: The job ID returned by hire_agent for long-running agents.

Embed and store text in a named collection for later similarity search.

Embeddings use your configured keys in order: Voyage → OpenAI → Gemini → local hash, unless overridden via provider/model/dimensions params or EMBEDDING_* env vars.

Args: collection: Logical bucket name (e.g. "project-docs", "kb"). text: Full text to embed and store. doc_id: Optional stable id; a UUID is generated if omitted. metadata_json: Optional JSON object string (e.g. {"source":"readme.md"}). provider: Embedding provider override (openai, voyage, gemini, local). Empty = auto-detect. model: Embedding model override (e.g. "text-embedding-3-large"). Empty = provider default. dimensions: Output dimensions override (integer as string). Empty = provider default.

Semantic search over a collection using the same embedding route as vector_store_add.

By default, the query embedding is auto-matched to the collection's stored provider/model. Use provider/model/dimensions to override.

Args: collection: Collection name used with vector_store_add. query: Natural-language query. limit: Max results (1–50). provider: Embedding provider override for the query (openai, voyage, gemini, local). Empty = auto-detect from collection. model: Embedding model override (e.g. "voyage-code-3"). Empty = auto-detect from collection. dimensions: Output dimensions override (integer as string). Empty = auto-detect.

List all collections, or list documents in one collection.

Args: collection: If empty, list collection names and counts. Otherwise list docs in that collection. limit: Max rows when listing documents in a collection.

Ask a question about documents in a collection. Retrieves the most relevant chunks via semantic search, then uses an LLM to answer based on that context.

Works with any collection — those created by the embeddings-agent (hire_agent) or manually via vector_store_add.

Args: collection: Collection name (from embeddings-agent output or vector_store_add). question: Natural-language question about the stored data. limit: Number of context chunks to retrieve (1–20, default 5). provider: Embedding provider override for the query (openai, voyage, gemini, local). Empty = auto-detect from collection. model: Embedding model override. Empty = auto-detect from collection. dimensions: Output dimensions override (integer as string). Empty = auto-detect.

Search the marketplace for scripts matching a natural language query. Returns top 5 results with vending codes and descriptions.

Args: query: Natural language description of what you need (e.g. "analyze CSV data").

Run a marketplace script by its vending code (e.g. S-7K2M). Returns a job_id for polling via check_job.

Args: vending_code: The script's vending code from search_scripts (e.g. "S-7K2M"). user_input: Text input for the script (check the script's input_schema for format).

Get full details about a script: description, validation report, usage stats, similar scripts, and creator info.

Args: vending_code: The script's vending code (e.g. "S-7K2M").

Run code in a secure E2B sandbox. Returns stdout, stderr, and any errors.

Args: code: The source code to execute. language: "python" or "javascript" (default: python). packages: Comma-separated list of packages to install (e.g. "pandas,numpy").

Full security audit pipeline: Bug Hunter finds vulnerabilities, Test Goblin generates regression tests, Code Gremlin writes fixes. Returns combined report.

Args: code: The source code to audit (paste the full code).