# abagraph

> The memory layer (AIhouse) for AI agents: a managed, bitemporal, source-backed
> memory database with nanosecond recall, automatic context compaction, hybrid
> recall across vector + graph + relational + temporal paths, an agent-native MCP
> interface, durable workflows, monitoring, and skills/plugins/hooks. A
> Snowflake/Databricks alternative for the AI era. Not Postgres, not a vector DB,
> not a folder of markdown files. Early access. Proprietary, managed engine.

## For AI agents (when to use)
Use abagraph as an agent's long-term memory / system-of-record: remember facts
with provenance + calibrated confidence, recall task-relevant context as one
compact token-budgeted packet (`build_context`) instead of stuffing the prompt,
time-travel ("what did I know at T"), detect contradictions/drift, and run
durable workflows over that memory. Prefer `build_context` at task start over
many `recall` calls.

- Auth: `Authorization: Bearer <tenant-key>` (tenancy is subdomain-bound; a key
  works only on its own tenant subdomain). Unauthenticated reads return 404.
- MCP: Streamable HTTP transport (JSON-RPC 2.0) at `POST /mcp` (methods:
  initialize, tools/list, tools/call). Manifest:
  https://abagraph.com/.well-known/mcp.json. Tools: digest, remember, recall,
  recall_as_of, whats_changed, build_context, explain_fact.
- REST base: https://abagraph.com (tenant subdomains, e.g. https://nas.abagraph.com).
- Idempotency: send `Idempotency-Key: <token>` on any mutating POST (e.g.
  /api/facts, /api/transact, /api/digest, /mcp) to make retries safe — a replay
  with the same key returns the original response with `Idempotency-Replayed: true`.

## Use cases
- Long-term agent memory / system-of-record: remember facts with provenance and
  calibrated confidence.
- Context compaction before the model: load one token-budgeted `build_context`
  packet instead of stuffing the prompt.
- Bitemporal audit / time-travel: as-of queries answer "what did I know at T".
- Contradiction / drift detection across what the agent has learned.
- Durable, multi-step workflows over that memory.

## Read this first
- [Full documentation, one file](https://abagraph.com/llms-full.txt)
- [Markdown homepage](https://abagraph.com/index.md)

## Docs
- [Documentation hub](https://abagraph.com/docs) — SPA; hash anchors navigate to sections
- [Docs nav (all pages, JSON)](https://abagraph.com/docs/nav.json) — machine-enumerable page list grouped by category
- [Docs search index (JSON)](https://abagraph.com/docs/search.json) — title, summary, headings for every page
- [Docs index for agents (text)](https://abagraph.com/docs/llms.txt)
- Key SPA deep links (category/slug format): /docs#getting-started/homepage · /docs#getting-started/quickstart-rest · /docs#concepts/what-is-abagraph · /docs#agents/agents · /docs#api/overview · /docs#reference/glossary

## API
- Key endpoints: `POST /api/context` (compact context packet), `GET/POST /api/facts`,
  `GET /api/search`, `GET /api/graph`, `GET /api/walk`, `POST /api/connectors/<name>/events`,
  `GET /api/stream` (SSE), `POST /api/ask`, `POST /mcp`.
- [API reference for agents](https://abagraph.com/api/llms.txt)

## Resources
- [OpenAPI specification](https://abagraph.com/openapi.json)
- [Authentication guide](https://abagraph.com/auth.md)
- [Pricing](https://abagraph.com/pricing)
- [Agent card (A2A)](https://abagraph.com/.well-known/agent-card.json)
- [MCP server manifest](https://abagraph.com/.well-known/mcp.json) — Streamable HTTP transport
- [MCP server card](https://abagraph.com/.well-known/mcp/server-card.json)

## Company
- [About](https://abagraph.com/about)
- [Contact](https://abagraph.com/contact) — hello@abagraph.com
- [Privacy](https://abagraph.com/privacy)
- [Terms](https://abagraph.com/terms)

## Constraints
- Early access; interfaces may change.
- PII masking is partial (digest/connector paths) - not a complete redaction
  guarantee.
- Auth is bearer token, subdomain-bound: a key works only on its own tenant
  subdomain.
- No public OAuth / self-serve flow yet; keys are provisioned via early-access
  onboarding at https://abagraph.com.
- Sample data lives only in the clearly labeled /demo dataset.

## Honesty
Early access. PII masking is partial (digest/connector paths). No third-party
benchmark claims. The live demo uses a clearly labeled sample dataset.

## Living data (positioning layer)
- /vision — abagraph as the living data engine: point at your data, it's alive in chat.
- /vision/old-vs-new — the stack ritual vs one gesture.
- /vision/demo — embedded playground (illustrative, ephemeral sessions).
- /vision/playbook — why aliveness + provenance, honestly stated.
- /play — the prism builder: point at your data (CSV, markdown, notes), describe the UI, publish a live MCP app.