AI Agents

• Area: Intelligence
• Path: ai/agents
• Kind: Autonomous agent runtime (REST + WebSocket API)
• Version: v0.2.1
• Status: Active

Role in the stack

AI Agents is the execution layer for autonomous AI agents within the Koder stack. It exposes a REST + WebSocket API that other Koder products can call to run multi-step reasoning tasks, decompose complex goals into parallel swarms, and connect to external tools via MCP.

Key capabilities added in v0.2.1:

• Context Compression — long sessions summarise their own history, keeping sessions coherent beyond the model's context window
• Dynamic Swarm Planner — LLM-driven goal decomposition into a parallel task DAG, executed via the worker pool
• Federated Agent Registry — external agents (any model, any device) register and participate in swarms

Primary couplings

Public interfaces

REST

• POST /v1/agents — create agent
• POST /v1/agents/:id/runs — submit run (strategy: react / chain-of-thought / plan-and-execute)
• GET /v1/agents/:id/runs/:runId — poll status / result
• POST /v1/swarm/runs — dynamic swarm execution (goal → DAG → parallel)
• POST /v1/swarm/plan — inspect DAG without executing
• POST /v1/registry/agents — register external peer
• POST /v1/registry/agents/:id/heartbeat — liveness signal

WebSocket

• WS /v1/agents/:id/runs/:runId/stream — real-time step events (think / act / observe / completed)

Config (env)