Safe, reproducible, API-identical mock environments so your agents can learn, fail, and improve — without deleting real emails or booking phantom meetings.
Real APIs are unforgiving. One bad function call and your agent has archived your entire inbox, double-booked your Tuesday, or replied-all with hallucinated contract terms.
smolclaw gives agents a safe place to fail. Every environment is API-identical, fully seedable, and instantly resettable — so you can run thousands of trials without consequences.
Each environment is a complete mock of a real API — same endpoints, same schemas, same auth patterns. Swap one URL and your agent can't tell the difference.
Full Gmail API mock. 54 endpoints including messages, threads, labels, drafts, and history. FastAPI + SQLite with seedable scenarios.
Google Calendar API mock. Events, recurring schedules, attendees, free/busy queries. Test scheduling agents without real calendars.
Google Drive API mock. Files, folders, permissions, sharing. Let agents manage documents without touching real data.
Slack API mock. Channels, messages, threads, reactions. Test communication agents in isolated workspaces.
Three concepts. Drop-in replacement for real APIs.
Pick a scenario — default, long_context, or a custom task — and populate the environment with realistic data.
Point your agent at the mock API. Same endpoints, same auth, same response schemas. Your agent code doesn't change at all.
Snapshot diffs show exactly what changed. Deterministic verifiers check if the agent did the right thing. Reset and repeat.
Want the full optimization loop? envdash connects smolclaw environments with GEPA skill optimization, Harbor task runners, and automated evaluation pipelines.
Same endpoints, same request/response schemas, same error codes. Change one environment variable to switch between mock and production.
Deterministic scenarios from 57 emails to 3,000+ complex threads. Content library generates realistic, diverse data every time.
Save state, run an agent, diff the results, restore. Run thousands of experiments from the exact same starting point.
Every environment ships with Dockerfile and docker-compose. One command to spin up isolated, reproducible test environments.
Separate admin endpoints for seeding, snapshots, state inspection, and task management. Full control over the environment.
MIT licensed. Contribute environments, seed scenarios, and evaluation tasks. Built by the community, for the community.
Up and running in under a minute.