Glossary
The words of Space Context
Space Context is the context layer for the physical world: film a space once and it becomes a structured Space Document that people and AI agents can read.
These are the words the API, the docs and the console use, one definition each.
- Agent Skill
-
A folder that teaches an AI agent one space: a static SKILL.md with the rules for reading it, space.json, and reference files for the areas, items, key locations, how-to steps, issues, facts, Q&A and frames. GET /v1/spaces/{id}/skill returns it; the CLI writes it with --format skill.
- Area
-
A room or part of a space: a kitchen, the cab of a camper, the engine room of a boat. Each area has a type, a condition, a description and an orientation sentence that fixes what left and right mean in it.
- Capture profile
-
What to film for one type of space: its key locations, each with a one-line hint and whether it is usually there, and its areas. There is one for each of the 11 types of space, in 7 languages, at GET /v1/capture-profiles/{space_type}, with no key needed.
- Check-in and check-out comparison
-
The difference between two scans of the same space: items added, moved or changed, and new issues, each with the frames of both scans. Something counts as removed or resolved only when a frame of the later scan shows it; everything else the later scan did not show is "not seen". GET /v1/spaces/{id}/diff.
- Coverage
-
What a scan could not establish, said instead of guessed: warnings about the footage, what a reader would expect but the recording did not document, and the checklist of key locations for the type of space, each seen or not seen.
- Enrichment
-
The stage of a scan that turns what was seen and said into items, issues, key locations, instructions, facts and Q&A, and marks sensitive values. enrichment.status says whether a document has been through it; until then a key without spaces:sensitive sees its shape and counts only.
- Evidence frame
-
A frame of the original video or photos that a record cites, with its time in the video. Frames are never retouched; their signed URLs expire, so records point to them by id.
- Frame check
-
POST /v1/frames/check: in about half a second, whether one frame is usable, which area it most likely shows and which key locations of the type of space it contains. Free within a daily allowance per account; nothing is stored.
- Guided capture
-
Using a capture profile and frame checks while someone films, so that an app can tick off key locations as they appear and point at what is still missing before the recording stops.
- Issue
-
A problem a scan saw or was told about (damage, wear, a fault, something missing), with its category, severity, suggested action and the frames that show it.
- Item
-
One thing in a space: its name and generic name, category, quantity, condition, brand and model when they are readable, visible text, aliases for search, where it is, and the frames it was seen in.
- Key location
-
Where the things people ask about first are: the fuse box, the water shut-off, the boiler, the router, a VIN plate, the seacocks of a boat. Each type of space has its own list, and each key location is a record with its photo.
- MCP server
-
Space Context’s remote Model Context Protocol server (Streamable HTTP, OAuth). It lets assistants such as Claude, ChatGPT, Claude Code, Codex, Cursor or VS Code read the spaces a person filmed, and scan new ones when the person allows it.
- Photo check (verification)
-
After a scan, every item, issue and key location is asked again, as a closed question, on the frames it cites, by a second model that cannot write text. Each record comes back verified, unconfirmed or not_visual, with the probability and the frame that showed it best.
- Physical-world prompt injection
-
Text in a space (a sign, a note, a sentence said on camera) that tries to give orders to the AI that later reads the space. Space Context withholds it from documents, skills, search and answers, and notes it in coverage.warnings.
- Placement
-
Where an item is, described the way a person would: an anchor (a wall, a door, another item), a relation such as on, inside, left of or behind, and the viewpoint that makes left and right true.
- Report language
-
The language a scan’s document is written in: English, Italian, Spanish, French, German, Portuguese or Dutch. The narration can be in any language.
- Scan
-
One recording of a space, a video of up to 10 minutes and 4 GB or up to 120 photos, turned into a new version of its Space Document. A live scan costs $1.00 from prepaid credit after a free first one with a card on file; a failed scan is refunded.
- Space
-
A place you document and come back to: a home, a rental, an office, a shop, a warehouse, a garage, a vehicle, a camper, a boat or an appliance. Each scan of it adds a version, and two versions can be compared.
- Space Document
-
The structured, cited JSON a space becomes: areas, items, issues, key locations, instructions, facts, Q&A, evidence frames and coverage. It is the public contract of the API, and the same document comes as Markdown and as an Agent Skill.
- spaces:sensitive
-
The API key scope that allows reading sensitive values such as door codes, Wi-Fi passwords and identification numbers. Keys are created without it: a key given to an agent reads the same document with those values redacted.
- Test key and live key
-
A test key (sc_test_…) is free, runs the same flow as a live one and always returns the demo space; a live key (sc_live_…) reads and scans real spaces and spends credit. Test and live never see each other’s data.
- x402
-
An HTTP payment protocol. An AI agent without an account pays in USDC on Base or Solana, in the request itself, and receives an API key and credit: POST /v1/agent/credits.
See it all in one real document on the demo page, or get a free test key.