Built for the way you work.
One API controls the full agent lifecycle: create a run, subscribe to logs, inspect files, write follow-up instructions, and manage the connected Git repository.
Multi-agent mode runs BLACKBOX, Claude Code, Codex, and Gemini against the same task in parallel, so your team compares implementations instead of waiting on one attempt.
Every successful run can become a reviewable pull request with a scoped branch, clean commits, a generated summary, and links back to the run history.
What you get.
Programmatic task control
Create, observe, and manage agent runs over HTTP. Stream live logs, read and write files in the sandbox, and continue a conversation on an existing task.
Task status and the event stream are identical across agent runtimes. A Codex task and a Claude task move through the same states and send the same events, so your integration does not change when you change runtimes.
Browser-based execution
Start coding agents directly from your browser. No local setup, and no CLI installation. Connect a repository, describe the task, and the agent runs in the cloud.
A secure sandbox starts for each task, with full terminal, file system, and Git access. Agents can install dependencies, run tests, and build projects.
Multi-agent orchestration
Send the same task to BLACKBOX, Claude Code, Codex, and Gemini at the same time. Chairman LLM evaluates each implementation and selects the best result.
Monitor all agents in one dashboard. Compare the implementations side by side. You can override the selection from Chairman LLM.
Automatic PR management
Agents create pull requests with clean commits, clear titles, and detailed descriptions. Review diffs, request changes, or merge — all from the dashboard.
The API integrates with GitHub, GitLab, and Bitbucket. You can configure branch names, commit message formats, and review workflows.
One task can become four independent attempts and one clean PR.
The API is built for automation loops, not only chat. Your system can launch work, watch the run stream, add context, inspect the sandbox, and promote the winning branch.
Create the run
Send the repository, branch policy, model set, and task brief. The same contract starts a single agent or a multi-agent race.
Stream the work
Terminal output, file edits, test results, and agent notes arrive as ordered events that your UI or CI system can render live.
Compare implementations
Chairman LLM reviews the competing branches for correctness, scope, risk, and clarity. Humans can still override the decision.
Open review
The chosen result lands as a scoped pull request with a run summary, changed files, test evidence, and a branch your team can inspect.
Built for repeatable engineering operations.
Agent work has to fit the systems teams already trust: Git, CI, review queues, audit trails, and permission boundaries.
API-native control
Run creation, event streams, file reads, file writes, and follow-up prompts are exposed as product APIs so internal tools can orchestrate agents directly.
Parallel model selection
Use different agents for the same task when correctness matters, or route specialized work to the agent that performs best on that class of change.
Review history
Every run keeps the request, decision, and PR trail together, so teams can explain why an implementation was selected before it merged.
Common questions.
DO I NEED TO INSTALL ANYTHING?
No. The Agents API is plain HTTP — you can call it from any tool that can make a request. If you prefer not to write the call, the same agents run fully in the browser at app.blackbox.ai. Sign in with Google or GitHub, connect a repository, and start tasks.
WHAT GIT PROVIDERS ARE SUPPORTED?
The API supports GitHub, GitLab, and Bitbucket, with automatic PR creation and branch management. Enterprise supports self-managed instances (GitHub Enterprise Server, GitLab self-managed, Bitbucket Data Center) with SSO integration.
CAN I RUN MULTIPLE AGENTS AT ONCE?
Yes. You can run many agents at the same time on different tasks, or use Multi-Agent mode to send the same task to Blackbox, Claude Code, Codex, and Gemini in parallel. Chairman LLM evaluates each implementation and selects the best result.
HOW ARE AGENTS SANDBOXED?
Each task runs in an isolated cloud sandbox with its own file system, network, and process boundaries. Agents can install dependencies, run tests, and build projects inside the sandbox, with no effect on other tasks.
WHICH AI MODELS CAN I USE?
The Agents API supports frontier and open-source models that include Claude Opus-4.6, GPT-5.2, Gemini-3, Grok-4, Llama, and Mistral. You select the model and provider before each task, and you can use different models for different tasks.
HOW IS THE AGENTS API LICENSED?
The Agents API is included with an Enterprise agreement — per-token commits that decrease as your teams consume them, with concurrency limits set for the organization, not per seat. Contact sales to scope an agreement.
HOW IS MY CODE KEPT SECURE?
We provide end-to-end encryption and encrypt data at rest with AES-256. Enterprise adds a single-tenant deployment that is isolated from every other customer.