Skip to main content
Vima onboarding is straightforward now. There are three clean entry points:

Humans

Start with the landing page and dashboard.

Judges

Read the submission story and headline results.

Agents

Connect through the hosted MCP endpoint.

Shell Automation

Use the portable vima-agent CLI.

​
Human Path

Open https://vimaspatial.tech, then jump to the dashboard. The site shows the proof chain, live CII numbers, and agent handoff commands.

​
Agent Path

Preferred: 
https://vimaspatial.tech/mcp
Fallback: 
uvx --from "git+https://github.com/philip-chen6/vima.git#subdirectory=packages/vima-agent" vima doctor

​
Current Rough Edges

The CLI is not published to PyPI yet, so uvx vima-agent@latest is still a future publishing target.
  • The hosted MCP service is live; use /mcp/health for simple checks because /mcp itself expects a streamable HTTP MCP client.
  • GET /api/demo is local-video-backed and returns video_unavailable on production. Use /api/cii/frames, /api/spatial/zones, and /api/eval for judge-safe hosted evidence.
  • POST /api/temporal/run runs live reasoning and persists the result read by /api/eval, so prefer GET /api/eval unless you intentionally want a fresh live run.
  • Generated demo artifacts are hackathon artifacts, not benchmark-grade eval data.
  • Mintlify deploys from stephenhungg/vima-docs; the main repo sync workflow requires VIMA_DOCS_SYNC_TOKEN before push-to-docs automation works.
None of these block onboarding. A new agent can verify production, inspect CII rows, query spatial zones, and analyze sample frames without cloning the full backend.