Skip to content

Services

Services are the capabilities available to agents at runtime. Agents declare their service requirements in agent.toml and the supervisor routes requests to the appropriate worker via the message queue.

Available Services

Service Queue name Description
Inference infer Text generation via LLM
Embedding embed Vector embedding generation
Object Storage storage Persistent key-value object storage
Vector Storage storage Similarity search over vector embeddings

Inference

Generates text from a prompt using a registered model.

Backends:

Backend Engine Description
Ollama ollama Local inference via Ollama server
OpenRouter openrouter Remote inference via OpenRouter API

Parameters:

Parameter Type Default Description
max_tokens integer 1024 Maximum tokens to generate
temperature float 0.7 Sampling temperature

Embedding

Generates vector embeddings for text input, used for semantic search via vector storage.

Backends:

Backend Engine Description
Ollama ollama Local embedding via Ollama server

Parameters:

Parameter Type Default Description
dimensions integer 768 Embedding vector dimensions

Object Storage

Persistent key-value storage for agent data.

Backend: SQLite — file-backed, durable, content-addressed.

Vector Storage

Similarity search over vector embeddings, powered by sqlite-vec (MIT licensed).

Backend: SQLite via sqlite-vec.

Message Flow

All service interactions use typed messages on the queue:

  1. Agent sends a RequestMessage to the service queue
  2. Worker processes the request and sends a ResponseMessage back

See Queue System for the full message flow.

See Also