Skip to content

fmonsalvebravo-dev/orchestrion

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

20 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Orchestrion

Orchestrion is an agent-native task orchestration API designed for autonomous workers and AI agents.

It manages task lifecycle, worker leases, retry scheduling, dead-letter queues, and provides machine-readable recovery guidance on every response.


Why Orchestrion

Agents and distributed workers need a reliable way to coordinate task execution. Existing job queues are designed for human-written code with try/catch blocks and retry libraries. Agents need something different:

  • Safe concurrent claiming. Multiple workers can race for tasks without double-processing. FOR UPDATE SKIP LOCKED guarantees each task is assigned to exactly one worker.
  • Automatic recovery. If a worker crashes, its lease expires and the task is recovered automatically. No operator intervention required.
  • Predictable retries. Failed tasks return to the queue with a scheduled retry window, preventing retry storms and giving transient failures time to resolve.
  • Machine-readable recovery. Every response includes an agent_contract with next_actions that tells the agent exactly what to do next. No status code parsing required.

Key Features

  • Deterministic task lifecycle (pending → claimed → completed / dead_letter / cancelled)
  • Lease-based worker ownership with heartbeat renewal
  • Retry scheduling with scheduled_at delays
  • Dead-letter handling with explicit requeue
  • Idempotent task creation via Idempotency-Key
  • agent_contract on every response — machine-readable next actions and recovery guidance
  • MCP-compatible tool manifest (GET /v1/tool)
  • OpenAPI 3.1 schema (GET /v1/schema)
  • Agent discovery document (GET /.well-known/agent.json)
  • Plan-based billing with quota and rate limit enforcement

Example Workflow

# 1. Register an API key
curl -X POST https://api.orchestrion.dev/v1/keys/register

# 2. Create a task
curl -X POST https://api.orchestrion.dev/v1/tasks \
  -H "Authorization: Bearer orch_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "type": "process_document", "payload": { "url": "https://..." } }'

# 3. Claim a task
curl -X POST https://api.orchestrion.dev/v1/tasks/claim \
  -H "Authorization: Bearer orch_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "type": "process_document" }'

# 4. Send heartbeats while working
curl -X POST https://api.orchestrion.dev/v1/tasks/{id}/heartbeat \
  -H "Authorization: Bearer orch_live_..."

# 5. Complete the task
curl -X POST https://api.orchestrion.dev/v1/tasks/{id}/complete \
  -H "Authorization: Bearer orch_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "result": { "pages": 42 } }'

Every response includes agent_contract.next_actions telling the worker what to do next.


Agent Discovery

An agent with zero prior knowledge can discover the full API from a single URL:

GET /.well-known/agent.json

This returns links to the capabilities document, tool manifest, OpenAPI schema, and registration endpoint. The agent reads these once and understands every rule of the system.


Documentation

Document Description
AGENTS.md Agent integration guide
AGENT_CONTRACT.md agent_contract protocol specification
AGENT_WALKTHROUGH.md Step-by-step agent interaction walkthrough
ARCHITECTURE.md Internal system architecture
DESIGN_DECISIONS.md Key architectural decisions and rationale
CONCURRENCY_MODEL.md Concurrency safety model
RELIABILITY.md Reliability guarantees
OPERATIONS.md Production operations guide
SECURITY.md Security policy
SECURITY_MODEL.md Security model deep-dive
API_PHILOSOPHY.md API design philosophy
WHY_THIS_EXISTS.md Design motivation and problem statement
PROTOCOL.md Formal protocol specification
TESTING.md Testing methodology
LOAD_TESTING.md Load and performance testing
TEST_ENVIRONMENT.md Test environment description

License

All rights reserved.

About

Agent-native task orchestration API with deterministic lifecycle, lease-based workers, and machine-readable recovery via agent_contract.

Topics

Resources

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors