Orchestration

Run multiple agents together — sequential, parallel, router, hierarchical, and debate strategies with a shared blackboard and handoffs.

Cortex orchestration coordinates multiple agents working together. An orchestration is a stored config — a strategy plus the participant agents and their settings — that you run by name. During a run, agents share a blackboard (roster + prior contributions) and pass work through recorded handoffs.

Strategies

Strategy	Behavior
`sequential`	Agents run in order; each sees the prior outputs and the final answer is the last agent's.
`parallel`	All agents run concurrently on the same input; an optional aggregator agent synthesizes one answer.
`router`	A static rule map or a router agent picks one agent to handle the input.
`hierarchical`	A manager agent delegates subtasks to workers (JSON plan), then synthesizes the result.
`debate`	Debater agents argue over N rounds, then a judge agent renders a verdict.

Awareness and communication

Every participant sees a roster of the others (name + role) and the running log of contributions via the blackboard snapshot prepended to its input. When one agent hands off to another, an AgentHandoff plugin hook fires (and is recorded on the run), so audit and metrics extensions observe the collaboration.

Settings

{
  "max_concurrency": 4,
  "rounds": 2,
  "manager": "lead",
  "judge": "referee",
  "aggregator": "summarizer",
  "router_agent": "dispatcher",
  "router_rules": { "refund": "billing", "bug": "support" },
  "model": "smart"
}

Each strategy reads only the settings it needs; the rest are ignored.

Create and run over HTTP

# Create a debate orchestration
curl -X POST /v1/orchestrations -H 'Content-Type: application/json' -d '{
  "name": "should-we-ship",
  "strategy": "debate",
  "participants": [
    {"agent_name": "optimist", "role": "debater"},
    {"agent_name": "skeptic", "role": "debater"},
    {"agent_name": "referee", "role": "judge"}
  ],
  "settings": {"rounds": 2, "judge": "referee"}
}'

# Run it
curl -X POST /v1/orchestrations/should-we-ship/run \
  -H 'Content-Type: application/json' \
  -d '{"input": "Should we ship the feature this week?"}'

The run returns a summary:

{
  "run_id": "orch_01...",
  "status": "completed",
  "strategy": "debate",
  "output": "...the judge's verdict...",
  "duration_ms": 4213
}

Run history is available at GET /v1/orchestration-runs and GET /v1/orchestration-runs/:id.

In Go

// Persist a config once...
cfg := &orchestration.Config{
    ID:       id.NewOrchestrationConfigID(),
    Name:     "research-team",
    AppID:    "my-app",
    Strategy: orchestration.StrategySequential,
    Participants: []orchestration.Participant{
        {AgentName: "researcher", Role: "worker"},
        {AgentName: "writer", Role: "worker"},
    },
}
_ = eng.CreateOrchestration(ctx, cfg)

// ...then run it by name.
run, err := eng.RunOrchestration(ctx, "my-app", "research-team", "Summarize CRDTs")