converge¶
A distributed operating system for agents.
Converge provides the foundational substrate for multi-agent computation: identity, messaging, discovery, coordination, and execution so that networks of autonomous agents can operate without central control. The library is agent-first, network-native, and task-oriented; it exposes primitives and pluggable transports, policy, and observability rather than a single prescribed workflow.
Documentation
Features¶
Agent-first: Identity, capabilities, and execution loop are first-class; agents are the only actors.
Network-native: Agents are peers; no central controller is required.
Task-oriented: Discovery, execution, coordination, and evaluation are explicit and inspectable.
Observable: Message flows, task state, and decision points can be logged, traced, and replayed.
Transport-agnostic: Local in-process, TCP, or pluggable transports via a common interface.
Pluggable persistence: In-memory or file-backed stores for pools, tasks, and discovery; crypto and LLM extensions optional.
Documentation map¶
Section |
Contents |
|---|---|
User guide |
Installation, quick start, core concepts, extensions, CLI and configuration. |
API reference |
Full API for all public modules (core, network, coordination, runtime, policy, observability, extensions). |
Architecture |
Design principles, module layout, process model, failure handling. |
Quick links¶
Concepts: agents, identity, messages, topics, tasks, pools, decisions
Extensions: crypto, LLM, storage
API reference: Project reference (module index)
Requirements and status¶
Python: 3.11 or newer.
Status: Experimental; APIs may change. Use at your own risk.
License and contributing¶
See the project repository for license and contribution guidelines.