01
Send one request
Choose a runtime, model, prompt, tools, permission mode, and workspace source through one consistent API.
POST /agent_run
Agent execution infrastructure
One API. Every agent runtime.
Run coding and general-purpose agents across local sandboxes, your cloud, or managed infrastructure—without locking your product to one provider.
API-first Self-hostable Provider-flexible
POST /api/v1/agent_run
{
"runtime": "openai_agents_sdk",
"model": "gpt-5.5",
"prompt": "Fix the failing tests",
"repo_url": "github.com/acme/app",
"permission_mode": "acceptEdits"
}
Live execution
RUNNINGWorkspace clonedmain
Tests executed12 passed · 2 failed
Agent editingtests/auth.test.ts
One control plane across your agent stack
Claude Agent SDK
Local sandbox
OpenAI Agents SDK
Local sandbox
Deep Agents
Local sandbox
Cursor SDK
Local + cloud
Claude Managed Agents
Cloud + self-hosted
Custom images
Your runtime
How it works
From prompt to production output.
Agent Studio turns a complex, provider-specific execution stack into one predictable workflow your application can call.
02
Prepare the workspace
Start from a Git repository, archive, previous run, or empty workspace. Configuration is discovered automatically.
Workspace ready
03
Run in isolation
The selected agent executes inside a dedicated sandbox or managed environment with explicit resource and tool controls.
Secure execution
04
Stream the result
Receive live progress through webhooks, answer approvals, resume sessions, and download the finished workspace.
Events + artifacts
See it in action
Agents that work inside real projects.
Launch a run, follow its tool calls and progress, then collect the completed workspace through your product.
Agent Studio · Live run
Fix failing authentication tests
RUNNINGrun_7f26a9 · session_82bc
Run the test suite, diagnose the two authentication failures, implement the smallest safe fix, and report what changed.
Workspace ready
Cloned acme/web-app · main
Executed tests
12 passed, 2 failed in auth.test.ts
Inspected source
src/auth/session.ts · 86 lines
Applying fix
Updating expired session handling…
Live output
agent Identified stale token reuse after refresh.
edit src/auth/session.ts
test Re-running targeted authentication suite…
A complete execution layer
Control the work, not just the model.
Configure the environment, capabilities, permissions, context, and output lifecycle around every agent run.
Your infrastructure. Your policies.
Run locally or self-host, inject secrets only when needed, constrain tools, and keep every workspace isolated.
Runtime flexibility
Switch between leading agent SDKs without rebuilding your orchestration layer.
Workspace-aware
Load repositories, archives, or a prior run and keep every agent grounded in the right project context.
Human approvals
Choose plan, guarded, edit-friendly, or autonomous modes and answer tool requests programmatically.
Resumable sessions
Continue work with the same session and persistent runtime state instead of starting from scratch.
Live event delivery
Push in-progress messages and final results into your product with signed, retryable webhooks.
Portable outputs
Download complete workspaces and retain source changes, reports, artifacts, and execution history.
Production architecture
Built for dependable agent operations.
API servers accept and track runs while background workers orchestrate isolated sandboxes or managed environments. State, sessions, and events remain available throughout the lifecycle.
Per-run isolation
A dedicated container or managed session for each execution.
Durable run state
Persist status, messages, session IDs, inputs, and outputs.
Scoped secrets
Pass configuration and sensitive values only to the run that needs them.
Observable lifecycle
Track pending, running, completed, failed, and cancelled work.
Agent Studio API
Authenticate · submit · query · cancel
Execution control plane
QueueWorkersWebhooks
Local sandbox
Managed cloud
Self-hosted
Product resources
See how the platform fits your operation.
Explore FAQs, use cases, architecture, decision guidance, and a platform walkthrough for Agent Studio.
Build beyond a single provider
Put agents inside your product.
Tell us about your use case, infrastructure, and preferred runtimes. We’ll help you design a reliable path from prototype to production.
Talk to an agent engineer