# Track how facts about a person or account change over time, with dates and sources

> Track how facts about a person or account change over time, with dates and sources: 2 companies fully do it, 5 with a limit, 35 carry the tag. Every listing in Agent memory with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/agent-memory.temporal-graph/
- Markdown: https://www.anchorterminal.com/niches/agent-memory.temporal-graph/index.md (~1,700 tokens)
- Slim: https://www.anchorterminal.com/niches/agent-memory.temporal-graph/index.min.md (~280 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/agent-memory.temporal-graph/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Two or three companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 2 | 5 | 35 | 10 | 4 |

## What counts as done

A query for an entity returns the current fact and the superseded ones each with valid-from and valid-to dates and the source episode, and a time-filtered query returns only facts valid at that time.

How agents ask for it: did this buyer ever raise HOA fees; what medications was the patient on in March; when did this change and what was it before

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **past.dev (Revo Inc.)**: occurredAt and supersededAt on artifacts, occurredFrom and occurredTo windows, a queryTimestamp perspective, sources with excerpts, and trace lineage and timeline routes that say which memory replaced which.
- **Zep (Zep Software, Inc.)**: Facts are bi-temporal with created_at, valid_at, invalid_at and expired_at, episodes are stored verbatim as sources, edge searches take datetime filters, and CallSphere relied on the validity windows for healthcare records.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Basic Memory (Basic Memory LLC)**: Observations can carry dates and bm_search filters by valid_at, valid_overlaps and time_kind, but validity windows and supersession are written by hand and there is no entity graph with source episodes.
- **Graphicle (KarLex AI, Inc.)**: Reviewed memories carry validity, immutable revisions and source citations, queryable as of a time, but they are about a code project and must pass human review; not people or accounts.
- **Graphiti (Zep Software, Inc.)**: Edges carry bi-temporal metadata for point-in-time queries and get_episode_entities traces provenance, which CallSphere used for validity windows; but self-hosting needs Neo4j or FalkorDB and bulk writes hit LLM rate limits (issue 290).
- **Hindsight (Vectorize, Inc.)**: Facts carry occurred_start, occurred_end and mentioned_at and temporal search fills a date window, but there is no valid-to on a fact and invalidation is a manual PATCH.
- **MemPalace (MemPalace (open-source project))**: Triples carry valid_from and valid_to, supersede closes and opens at one boundary, query_entity takes as_of and timeline lists history in SQLite; but triples are written by the agent, not extracted, and carry no source episode.
- **Supermemory API + MCP (Supermemory)**: Updates, extends and derives relations with isLatest keep history and a date field backfills, but no valid-from or valid-to fields or point-in-time query appear in the docs read.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Memori (Memori Labs Inc.)**: A knowledge graph is documented, but the graph page was not read and no validity windows appear in the pages that were.

- not this job: Amazon Bedrock AgentCore Memory (Amazon Web Services, Inc. (regional AWS entities by account location)): Records carry createdAt and a timestamp but no valid-from, valid-to or graph; the episodic strategy stores episodes, not dated fact histories.
- not this job: AI Memory (Fabio Akita): Wiki pages with version chains; no dated facts with valid-from and valid-to about entities.
- not this job: claude-mem (Alex Newman): No validity windows or entity graph; observations carry dates only.
- not this job: Cognee (Topoteretes UG (haftungsbeschränkt)): The graph has no validity windows; temporal_columns exist only for structured dlt ingestion.
- not this job: Context Mode Engine (MKSF LTD): No facts or entities; the Engine stores tool output and session rows.
- not this job: Egoist AI Passport (Egoist Machines, Inc.): No dated facts, validity windows or graph in the agent docs.
- not this job: EverOS (Evermind AI, Inc.): Episodes and facts carry timestamps; no valid-from or valid-to windows or graph edges.
- not this job: Agent Platform Memory Bank (Google LLC): Revisions record change history but there are no valid-from and valid-to facts or a graph to query by time.
- not this job: Honcho (Plastic Labs, Inc.): Conclusions carry premises, not valid-from and valid-to windows, and there is no entity graph to query at a time.
- not this job: Hy Memory (Tencent): The graph layer holds schemas and mental models; only intentions carry an expiry, and no fact has valid-from and valid-to windows.
- not this job: LangMem (LangChain (the licence file gives no legal form)): Memories are store documents with updated_at; no validity windows or graph.
- not this job: Mem0 Platform + MCP (Embedchain, Inc. (DBA Mem0)): Graph Memory, timestamps and Temporal Reasoning order results by time, but ADD-only extraction never closes a fact, so no valid-to or superseded history exists.
- not this job: MemMachine (MemVerge, Inc.): Episodes carry timestamps and search sorts by them; no validity windows or point-in-time query.
- not this job: TencentDB Agent Memory (Tencent): Atoms carry created_at and updated_at and a version; no validity windows or point-in-time query.
- not this job: Tokenrip (Quintel, LLC): Artifact versions have dates; no dated facts about entities.

## What stops agents finishing

- self-hosting needs a graph database (Neo4j or FalkorDB) running
- each write runs LLM extraction, so a burst hits the provider's rate limit (Graphiti #290)
- MCP server needs OAuth through Google Workspace, so a headless agent can't use it (Zep)

## Around it

Before: crm.sync, document-extraction.records. After: agent-memory.user-facts-recall.

## Where this comes from

Weight 10 of 100 in Agent memory: 4 of 36 accounts (CallSphere on Zep, three Graphiti issues) plus Thoughtworks' Trial rating; fewer accounts than the vendor pages suggest

As data: https://www.anchorterminal.com/api/v1/use-cases/agent-memory.temporal-graph.json, or the MCP tool get_use_case with id agent-memory.temporal-graph. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
