v i m a.
Video intelligence for construction sites. No lidar.
Hardhat video becomes an inspectable evidence chain. A depth-delta pre-pass drops two thirds of low-signal frame pairs, COLMAP registers the rest into a sparse cloud, and episodic memory binds events to time and frame.
evidence chain · one scroll, one proof path
every payout starts as a frame you can inspect.
vima turns egocentric video into timestamped work claims, then binds those claims to spatial zones before settlement logic sees them.
each bodycam frame gets paired with model depth before it can become a spatial work claim.
left: hardhat masonry frame. right: per-frame depth map used before spatial memory.
frame stream · drag to inspect
episode taxonomy · five shapes from the paper
every claim is one of five shapes.
Vima's episodic memory emits one of five episode types per detected event. The shape makes the claim reviewable in seconds. A human swipes, the claim resolves, the reviewed evidence enters the audit trail.
masonry_work_candidate — observed work in progress on the masonry plane
scaffold_zone_visible — scaffolding plank or rail in the active frame
safety_edge_context — exposed edge or fall hazard in view
foreground_worker_present — primary subject in motion or labor
ledger · settlement receipt
this is where a frame becomes inspectable evidence.
the ledger is the audit handoff: frame labels stay visible while the payout gate scores productive time, blocks idle work, and preserves the receipt.
verify · human review of episodic memory claims
every claim earns its truth.
the shipped ios verifier turns spatial claims into a swipe queue. one gesture confirms, rejects, or skips; the session tracks xp, streaks, raffle progress, and payout eligibility.
pipeline
video in. auditable work claims out.
agent handoff · hosted api + mcp + cli + skill
plug vima into any agent.
the dashboard is not the only interface. vima now exposes a hosted mcp server for claude code, cursor, windsurf, and other tool-using agents, plus a thin cli around the same production api for stable json, frame analysis, cii receipts, temporal evals, and skill handoff prompts.https://vimaspatial.tech/mcpuvx --from "git+https://github.com/philip-chen6/vima.git#subdirectory=packages/vima-agent" vima doctoruvx --from "git+https://github.com/philip-chen6/vima.git#subdirectory=packages/vima-agent" vima analyze --sample masonry-p --jsonuvx --from "git+https://github.com/philip-chen6/vima.git#subdirectory=packages/vima-agent" vima skill print --agent codex