# InvariantDB

> InvariantDB is a bitemporal graph database for AI agent memory and audit. It records when each fact was true and when the database stored it, keeps belief revisions with their evidence, and records the credential that made each write (from 1.4.2 for every write of an authenticated request except import jobs and the legacy /write route; on 1.4.1 and earlier when the request includes provenance). Agents connect through a Model Context Protocol (MCP) endpoint or use Cypher and REST over HTTPS.

InvariantDB uses Cypher with two time clauses: `AT VALID` (when a fact was true) and `AT RECORDED` (when the database stored it). Beliefs are nodes linked to their revisions and evidence. Writes can carry provenance (actor, source, reason, confidence) that the engine stores with the authenticated principal. From 1.4.2 a graph can carry an access policy by label and relationship type (`db.access.*`), a constraint can require a relationship (`CREATE CONSTRAINT ... REQUIRE (c)-[:T]->(:M)`), and HTTP write requests accept an `Idempotency-Key` header. From 1.4.2.2 a vector search can be narrowed by a graph pattern or a list of nodes. From 1.4.2.4 a query without a time clause answers what is valid now, and from 1.4.2.5 a backdated write that would replace later-dated versions is refused. From 1.4.3 read statements run on a planned query engine with the same rows, and `db.createFulltextIndex` declares a full-text index with phrases, Boolean and proximity connectors, snippets and reads at a past time (`db.fulltext.query`). From 1.4.3.1 memory recall also returns memories linked through saved facts, concurrent reads do not see a write statement before it commits, and filtered vector searches on large indexed labels may be approximate at the filtered recall target. From 1.4.3.2 the query planner is on by default (same rows; row order without `ORDER BY` can differ), `FOR PORTION OF VALID FROM ... TO ...` corrects a valid interval and keeps later versions, relationship properties can be indexed, `gds.vectorSearch` takes `rescore` (column `exact`), and a wrong-type `size()`, `properties()` or `keys()` answers `ARGUMENT_TYPE_MISMATCH`. From 1.4.3.3 counts over a label or the whole graph are not refused at the row budget, `db.fulltext.query` searches several fields with weights and takes `filter`, `matched_terms`, `excerpt` and `group_by`, `db.hybridSearch` takes `fusion`, `weights` and `at_valid` / `at_recorded` (with filters), `CREATE INDEX` on a large label builds in the background (`INDEX_POPULATING`), declared column files are opt-in (`db.createColumn`), new graphs store their history compressed, and `/graphs/{name}/nlq`, `/investigate` and `/agent` answer 410 `FEATURE_RETIRED`. `db.subjectExport` returns one subject's node, its relationships and its property versions (GDPR Article 20); it matches the node by its string id. Crypto-shred (per-subject encryption keys and key destruction) is in development. The API is Cypher and REST over HTTPS plus an MCP endpoint with 28 tools.

This file follows the [llmstxt.org](https://llmstxt.org/) convention. It is an index of public pages for agents, indexers and evaluation tools, and it does not link to admin or internal pages. `/llms-full.txt` has the full reference in one document.

## For AI systems

- [For AI overview](https://invariantdb.com/for-ai.html): Overview for AI agents, with links to the MCP endpoint and the Cypher and REST references.
- [Agents guide](https://invariantdb.com/agents.html): How to give an agent an auditable memory. Covers episode recording, belief revision and recall patterns.
- [AI memory concept](https://invariantdb.com/ai-memory.html): What agent memory means in a bitemporal graph: episodes, beliefs, salience and lineage.
- [Agent memory demo](https://invariantdb.com/agent-memory-demo.html): Static sample of an agent's memory (events, entities, facts with the time each was recorded); replay what it held at an earlier time.
- [Agent memory walkthrough](https://invariantdb.com/agent-memory-walkthrough.html): Self-running walkthrough of episodic memory, belief revision, salience decay and provenance.
- [Memory](https://invariantdb.com/memory.html): Set up a memory workspace and add it to an MCP client as a remote MCP server.

## Product capabilities

- [Bitemporal graph](https://invariantdb.com/product-bitemporal.html): Valid time and recorded time. `AT VALID` reads what was true in the world at time T; `AT RECORDED` reads what the database had stored at time T. Includes a timeline diagram and Cypher examples.
- [Provenance and belief lineage](https://invariantdb.com/product-provenance.html): `db.beliefLineage` returns `beliefId`, `claim`, `confidence`, `createdAt`, `supersededAt`, `depth` and `evidence` for each version of a belief, and `db.propertyProvenance` returns who changed a property value, from what source, and why.
- [Crypto-shred](https://invariantdb.com/product-crypto-shred.html): Crypto-shred (per-subject encryption keys and key destruction) is in development. On the current release, declaring an encrypted property is refused by default, and `db.destroySubjectKey` is refused without recording a receipt. Contact hello@invariantdb.com before relying on it.
- [Regulated workloads](https://invariantdb.com/regulated.html): Use in regulated industries, with material for regulations such as GDPR, HIPAA, NYDFS Part 500 and SOX.
- [Regulated walkthrough](https://invariantdb.com/regulated-walkthrough.html): Audit-response walkthroughs for 14 regulatory regimes.
- [Compliance dashboard](https://invariantdb.com/compliance.html): Audit-chain verification and compliance export for your graphs (Enterprise license).
- [Features](https://invariantdb.com/features.html): Catalog of shipped features.
- [Why "approved"](https://invariantdb.com/why-approved.html): Explainability case study that reproduces what an agent saw when it made a decision.

## Cypher and MCP

- [Cypher reference](https://invariantdb.com/cypher.html): Bitemporal Cypher, including the `AT VALID` / `AT RECORDED` clauses, `db.beliefLineage`, `db.recordEpisode`, `db.replayEpisodes`, `db.destroySubjectKey`, `db.proveErasure`, `db.verifyAclChain`, `db.subjectExport`, full-text indexes (`db.createFulltextIndex`, `db.fulltext.query`), and standard Cypher.
- [MCP connect](https://invariantdb.com/mcp.html): Hosted endpoint `https://invariantdb.com/graphs/{graph}/mcp` (one per graph; `POST /mcp` serves the default graph on self-hosted servers only), JSON-RPC 2.0 over HTTPS, authentication with a `gq_` API key or OAuth sign-in, the 26-tool catalog, and client configuration examples. The server negotiates MCP protocol version 2025-06-18, 2025-03-26 or 2024-11-05.
- [Quickstart](https://invariantdb.com/quickstart.html): Sign up, create a graph, connect an MCP client, and run a first bitemporal query.
- [30-minute onboarding](https://invariantdb.com/onboarding-30min.html): Step-by-step path from sign-up to a bitemporal query, agent memory, and MCP.
- [API keys](https://invariantdb.com/apikeys.html): Create and manage `gq_` API keys, which are sent as bearer tokens.
- [SDK](https://invariantdb.com/sdk.html): Python and Node client libraries, and REST usage for HTTP integrators.
- [Schema](https://invariantdb.com/schema.html): Declaring labels, edge types, property indexes and vector indexes.
- [Reference](https://invariantdb.com/reference.html): API reference index.
- [Architecture (public overview)](https://invariantdb.com/architecture.html): Capability-level architecture: the MCP endpoint and how bitemporal queries map to the API.

## Demos

The demos use fictional data and show static sample output; they do not connect to a server.

- [Decision demo](https://invariantdb.com/decision-demo.html): Reproduce an agent decision from recorded time.
- [Insurance fraud](https://invariantdb.com/insurance-demo.html): Ring detection with belief revision.
- [AML](https://invariantdb.com/aml-demo.html): Anti-money-laundering pattern discovery.
- [Cyber breach](https://invariantdb.com/cyber-demo.html): Incident forensics against a bitemporal graph.
- [Clinical trial](https://invariantdb.com/clinical-trial-demo.html): Consent and evidence lineage.
- [Supply chain](https://invariantdb.com/supply-chain-demo.html): Provenance from source to shelf.
- [Digital forensics](https://invariantdb.com/forensics-demo.html): Reconstructing what a system knew.
- [Mission readiness](https://invariantdb.com/mission-readiness-demo.html): Fictional flight-readiness review: what was on record at decision time versus what was valid then.

## Product context

- [Homepage](https://invariantdb.com/index.html): Home page.
- [Pricing](https://invariantdb.com/pricing.html)
- [Documentation index](https://invariantdb.com/docs.html): Top-level docs table of contents.
- [Knowledge](https://invariantdb.com/knowledge.html): Build a connected graph from your documents, keeping each document as a versioned source.
- [Signup](https://invariantdb.com/signup.html): Create an account.
- [Login](https://invariantdb.com/login.html): Sign in.

## Legal

- [Privacy policy](https://invariantdb.com/privacy.html)
- [Terms of service](https://invariantdb.com/terms.html)

## Optional

- [Full reference](https://invariantdb.com/llms-full.txt): Single-document reference for agents: Cypher dialect, procedure catalog, MCP tools, REST endpoints, authentication and common mistakes.

---

Version: 2026-10-17
Last updated: 2026-10-17
Canonical URL: https://invariantdb.com/llms.txt