For software factories

Run every coding workflow. One cloud-agent runtime.

Your factory decides the queue, policy, and budget. AgentSky runs each supported coding agent, model, and sandbox through the same API.

Talk to Us View Architecture

Agent · model · computer · state · events

Claude CodeCodexHermesOpenClaw+ more

WorkflowAgentModel

Review pull requestsCodexGPT-5.6 Sol

Refactor a serviceClaude CodeClaude Opus 5

Research dependenciesHermesDeepSeek V4 Pro

Operate a workflowOpenClawKimi K3

One clear boundary

Your factory routes. AgentSky runs.

Keep the workflow graph in the system that already understands your product. Call AgentSky whenever a step needs a coding agent with a real computer and durable state.

Your factory owns

Jobs, queues, and scheduling
Policies, approvals, and budgets
Tenants and product experience
Retries, results, and business logic

AgentSky runs

Supported agent and compatible model
Isolated sandbox
Session, workspace, and recovery
Messages, events, and runtime logs

Operate many workers

Scale the work without maintaining every runtime.

Create separate cloud sessions for independent jobs. Your orchestrator controls how much work enters the system; AgentSky gives accepted work one consistent execution contract.

Queue + concurrency

Your factory: Own priority, fan-out, global concurrency, and backpressure.

AgentSky: Accept asynchronous turns and provision sessions eagerly or on first work.

Tenant isolation

Your factory: Map customers, roles, repositories, and policy.

AgentSky: Scope API access by universe and isolate each session in its own workspace.

Budgets + cost

Your factory: Set limits by tenant, workflow, team, or job.

AgentSky: Meter runtime usage and stop new work when the prepaid balance is exhausted.

Audit + observability

Your factory: Record routing decisions, approvals, and business outcomes.

AgentSky: Expose session metadata, live events, message history, and runtime logs.

Reliability + recovery

Your factory: Decide job retries, fallbacks, and terminal failure policy.

AgentSky: Preserve the runtime volume across restarts and restore session context from backups after machine loss.

Fits your control plane

Add one agent API, not another orchestration system.

Attach your job and tenant references to session metadata, stream the work, and reconcile results in the system you already operate.

Read API Docs See Tycoon architecture

Your orchestratorAgentSky API

POST /api/v1/sessions
{
  "agent": "pull-request-reviewer",
  "metadata": {
    "tenantId": "acme",
    "workflowId": "pr-482"
  },
  "initial_events": [{
    "type": "user.message",
    "parts": [{ "type": "text", "index": 0,
      "text": "Review pull request #482" }]
  }]
}
GET /api/v1/sessions/{session_id}/stream
GET /api/v1/sessions/{session_id}/events

What AgentSky operates

A complete cloud worker behind every accepted job.

The integration stays the same across supported harnesses. Your factory can change the agent or model without rebuilding the worker lifecycle underneath it.

Compatible agent + model

Choose a supported pair for each kind of coding work.

Isolated sandbox

Each session gets its own runtime and persistent workspace.

One event contract

Send, stream, interrupt, inspect history, and read logs through the same interface.

Runtime recovery

AgentSky owns pod replacement and workspace restore; your factory owns business retries.

Your software factory. One cloud-agent runtime.

Tell us how your orchestrator routes work today. We’ll map it to the current AgentSky API and supported runtime catalog.

Talk to Us View Architecture