​
Agents

​
What is an Agent?

An agent is an autonomous AI process that:
  1. Receives a task description
  2. Gets its own isolated git worktree
  3. Reads and modifies code
  4. Runs tests and linters
  5. Commits and pushes its solution
  6. Waits for the Chairman to evaluate it

​
Agent Lifecycle

QUEUED → RUNNING → DONE / FAILED
                ↓
          CHAIRMAN_EVAL
                ↓
          WINNER_SELECTED → PR_OPENED

​
Multi-Agent Execution

When you run a task, SynapseX spawns N parallel agents (configurable, default 3), each using a different model: 
synapsex agent create bugfix-agent --model gpt-4o
synapsex agent invoke bugfix-agent "Fix the login bug"
For orchestrator-backed multi-agent runs, call the orchestrator proxy endpoints through the gateway or use the VM/Ops CLI workflows documented in the CLI reference.
AgentModelScore
agent-01azure/gpt-4-1-teacher0.94
agent-02azure/gpt-4o0.81
agent-03claude-sonnet-4-60.87
The Chairman selects agent-01 and opens a PR.

​
Supported Models

Any supported model can be selected when creating an agent with --model:
  • azure/gpt-4-1-teacher — best quality (default)
  • azure/gpt-4o — balanced
  • azure/gpt-4o-mini — fast, cheap
  • claude-sonnet-4-6 — Anthropic
  • gpt-4o — OpenAI direct