Choose a client
Both standalone SDKs support agents, builds, runs, skills, integrations, batches, recurring loops, campaigns, cancellation, and usage. They send authenticated requests to https://api.grep.ai/api/v2 by default. An API key is required for every request.
| Feature | TypeScript / JavaScript | Python |
|---|---|---|
| Distribution | @parcha/agentrun | parcha-agentrun |
| Import | AgentRun, Agent, Skill | AgentRun, Agent |
| Runtime | Node.js 20.9+ | Python 3.10+ |
| Runtime dependencies | None | None |
| Execution | Async promises | Synchronous |
| Polling units | Milliseconds | Seconds |
| Artifact downloads | Blob | bytes |
Version and availability
These pages describe SDK version 0.2.x and the unified v2 backend, which production serves at https://api.grep.ai/api/v2. Both packages are published: @parcha/agentrun on npm and parcha-agentrun on PyPI. Changing the base URL cannot add missing routes to an older deployment; confirm a non-production deployment against its own /api/v2/openapi.json.
Install with npm install @parcha/agentrun or python -m pip install parcha-agentrun, pinning the version your integration was reviewed against (0.2.0 or later). The Python import is agentrun, not the distribution name. Neither SDK falls back to legacy routes or retries writes automatically.
Authentication and permissions
Keep AGENTRUN_API_KEY in your server environment. Both clients send Authorization: Bearer <key>. Keys retain scope, ownership, and account feature checks: agent and skill management uses experts:read / experts:write, research workflows use research:read / research:write, and usage uses billing:read. Batch, campaign, and recurring access must also be enabled for the account.
Set baseURL in TypeScript or base_url in Python only to use another authorized Grep deployment. Remote URLs require HTTPS; localhost is allowed for development. Never put a secret key into browser code.