# abagraph — the memory layer for AI agents

abagraph is a managed, bitemporal, source-backed **memory database for AI agents** —
an "AIhouse", the system of record an agent reads, writes, reasons over, and
time-travels through. It unifies vector, graph, relational, and temporal recall in
one managed engine, reachable over an agent-native **MCP** interface and a REST API.
It is **not** Postgres, a vector DB, a graph DB, or a folder of markdown files; it
is a Snowflake/Databricks alternative for the AI era. Status: **early access**;
the engine is **proprietary** (select components may become source-available in the future).

## What makes it different
- **Bitemporal by default** — every claim carries valid-time and record-time. A
  contradiction *closes* the prior fact instead of overwriting it; the history is
  the data. Ask "what did I know at time T" with `as_of` queries.
- **Source-backed** — every fact has provenance and calibrated confidence;
  low-confidence facts are held back (abstention), never silently asserted.
- **Hybrid recall** — vector, graph, relational, and temporal access paths in one
  store; fused semantic + graph recall completes in well under a millisecond.
- **Automatic context compaction** — `build_context` returns a confidence-ranked,
  token-budgeted, fully-sourced packet to inject *before* your agent, so its
  context window never floods.
- **Operations built in** — durable workflows + DAGs, scheduled reports,
  monitoring/alerts on confidence/freshness/drift/cost, structured logs + traces,
  plus skills, plugins, and hooks. Multi-tenant isolation and an immutable audit trail.

## For AI agents
Use abagraph as long-term memory. Prefer `build_context` at task start over many
recall calls. Authenticate with `Authorization: Bearer <tenant-key>` (tenancy is
subdomain-bound). Call tools over MCP at `POST /mcp`.

```bash
# A compact, confidence-ranked, sourced context packet for a task
curl -s https://nas.abagraph.com/api/context \
  -H "Authorization: Bearer $ABAGRAPH_KEY" -H "Content-Type: application/json" \
  -d '{"seeds":["Creator:maya"],"goal":"plan the next launch","token_budget":80}'
```

## Links
- Full docs (one file): https://abagraph.com/llms-full.txt
- Agent index: https://abagraph.com/llms.txt
- Docs: https://abagraph.com/docs
- About: https://abagraph.com/about · Contact: hello@abagraph.com

_Honesty: early access; PII masking is partial; no third-party benchmark claims._