Introduction
Learn how to use and build with the coSPEC Platform
What is coSPEC?
coSPEC is API-first infrastructure for running AI coding agents securely. You send a prompt and a repository — coSPEC spins up an isolated sandbox, runs the agent, and returns the results.
Built for teams building automated workflows with tools like Make, Zapier, n8n, CI/CD pipelines, and your own applications.
How it Works
Set up an automation in Make, Zapier, n8n, a CI/CD pipeline, or your own application — then let coSPEC handle the execution:
Start a run — send a prompt, a repository, and optional configuration via a single API call.
coSPEC executes the agent — provisions an isolated gVisor sandbox, clones your repository, and launches the coding agent.
Get the output — text responses, branches, pull requests, or issues depending on your configuration.
Core Concepts
Runs
A run is a single agent execution. You send a prompt and a repository — coSPEC executes the agent and returns the output. Each run gets its own sandbox, controlled network access, and full audit trail.
Agents
An agent is the AI coding tool that executes inside the sandbox. Currently, coSPEC supports Claude Code — Anthropic's CLI coding agent. The agent receives your prompt, works with your repository, and produces the output. You provide your own Anthropic API key — coSPEC handles the infrastructure.
Sandboxes
Every run executes inside an ephemeral gVisor container. Sandboxes provide kernel-level isolation — no shared resources between runs or tenants. They are created automatically when a run starts and destroyed after completion.
Templates
Templates are reusable sandbox configurations. They define the base image, guardrails (cost and time limits), environment variables, and allowed egress domains. Set once, use across runs.
Next Steps
Get up and running with your first agent execution in under 5 minutes.