# Turn documents, exports and records into memory the agent can recall with citations

> Turn documents, exports and records into memory the agent can recall with citations: 3 companies fully do it, 15 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.ingest-records/
- Markdown: https://www.anchorterminal.com/niches/agent-memory.ingest-records/index.md (~2,000 tokens)
- Slim: https://www.anchorterminal.com/niches/agent-memory.ingest-records/index.min.md (~380 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/agent-memory.ingest-records/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 |
| --- | --- | --- | --- | --- |
| 3 | 15 | 35 | 8 | 4 |

## What counts as done

Ingest status can be polled, and after ingest completes a question answered from the records returns the passage with a source reference.

Measured, for our own tests: the count of stored items matches what was sent.

How agents ask for it: load my ChatGPT export; remember everything in these meeting notes; index the wiki so you can answer from it

## Does the job

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

- **Cognee (Topoteretes UG (haftungsbeschränkt))**: remember takes text or base64 files up to 10 MB, background=True returns at once with cognify_status and dataset status to poll, RememberResult lists items, and recall with include_references cites data_id and dataset_id (issue 633 asked for this).
- **Hindsight (Vectorize, Inc.)**: Batch retain and file retain run asynchronously with one operation_id per file, GET /operations/{id} reports status with retry and cancel, items_count is returned, and include_chunks returns the verbatim source.
- **past.dev (Revo Inc.)**: Batch ingest takes up to 1,000 data points and 16 MiB, GET /ingest/{id} reports processing, completed or failed with readiness counts, failed sends retry free, and recall cites sources with verbatim excerpts.

## 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.

- **Amazon Bedrock AgentCore Memory (Amazon Web Services, Inc. (regional AWS entities by account location))**: IngestData accepts conversational and non-conversational payloads and records appear in seconds to minutes, but there is no per-ingest status id or stored count; failures surface only through ListMemoryExtractionJobs and the LTM_RATE_EXCEEDED quota.
- **AI Memory (Fabio Akita)**: Imports one local PDF, DOCX, PPTX, XLSX, Markdown or transcript file at a time and memory_write_pages_batch writes up to 100 pages atomically, but there is no bulk ingest route, status or count for exports of thousands of chunks.
- **Basic Memory (Basic Memory LLC)**: PDFs, Word, PowerPoint and CSV files become searchable Markdown notes linked to the original and hits cite note paths, but no ingest status route or stored-count check was found for exports of thousands of chunks.
- **Context Mode Engine (MKSF LTD)**: ctx_index indexes a file, folder, text or URL into the project database and ctx_search returns the matching parts with their source label, but there is no status route, count check or scale figure for large exports.
- **EverOS (Evermind AI, Inc.)**: Knowledge documents have create, replace, patch, topic and search routes and ContentItem carries files, but no stored count or citation format was read, and strategy output can lag a flush by tens of seconds.
- **Agent Platform Memory Bank (Google LLC)**: Generation takes session history or pre-extracted memories to consolidate, runs asynchronously and can merge metadata, but no document ingestion, stored-count check or citation was found in the pages read.
- **Graphicle (KarLex AI, Inc.)**: Documents and sources are synced from folders and connectors and recalled as cited excerpts with byte offsets, but no bulk export ingest, status route or stored-count check beyond repository_status was found.
- **Graphiti (Zep Software, Inc.)**: add_episode_bulk loads text, message and JSON episodes, but 10,000 characters made over 1,000 OpenAI requests and failed on the rate limit with no record of what was written, and SEMAPHORE_LIMIT is the only throttle (issue 290).
- **Honcho (Plastic Labs, Inc.)**: File uploads turn PDFs, text and JSON into messages, a Mem0 migration imports conclusions, a queue status endpoint exists and chat answers expose evidence; stored counts and scale were not stated.
- **Mem0 Platform + MCP (Embedchain, Inc. (DBA Mem0))**: Direct Import seeds a project and infer=False stores raw text, with a PENDING event_id to poll; but OSS add() dropped memories whose embed call failed with only a warning (issue 5245), so counts can silently fall short.
- **MemMachine (MemVerge, Inc.)**: POST /memories adds a batch of messages with timestamps and metadata, but no document or export ingestion, status route or stored count was found.
- **MemPalace (MemPalace (open-source project))**: mine ingests files and conversations verbatim with no LLM, but the status count was capped at 10,000 against 27,387 drawers and a custom room was ignored (issue 599).
- **Supermemory API + MCP (Supermemory)**: Text, files up to 50 MB, URLs and connectors are ingested with queued, extracting, indexing, done and failed statuses and chunks for citations; but under dynamic dreaming memories can arrive later, and no prompt-injection guidance exists for ingested pages (map blocker).
- **TencentDB Agent Memory (Tencent)**: Wiki turns Markdown, Word, Excel, PowerPoint and PDF into linked pages and an import script loads past sessions, with processing status in the panel; no API status route or stored count was read.
- **Tokenrip (Quintel, LLC)**: artifact_upload, artifact_publish and tables from CSV store records and search finds readable artifacts, but there is no extraction, chunking or citation.
- **Zep (Zep Software, Inc.)**: zep-ingest and the Batch API load documents, transcripts and exports and graph.episode.add returns a processing task, but the batch status route was not read and each 350 bytes costs a credit.

- not this job: claude-mem (Alex Newman): Memory Ingest imports Claude Code auto-memory Markdown and export/import moves observations between installs; chat exports, PDFs and wikis are not ingested.
- not this job: Egoist AI Passport (Egoist Machines, Inc.): No ingest route for agents; connectors read live data such as calendar and mail at recall time.
- not this job: Hy Memory (Tencent): add() takes messages; no file, export or document ingestion route or status was found.
- not this job: LangMem (LangChain (the licence file gives no legal form)): Extraction runs over conversation messages; no document or export ingestion.
- not this job: Memori (Memori Labs Inc.): Memory is captured from wrapped LLM calls and agent traces; no document, export or file ingestion route is listed.

## What stops agents finishing

- 10,000 characters produced over 1,000 LLM calls and a rate-limit failure with no record of what landed (Graphiti #290)
- ingest status capped or wrong, so the agent can't tell whether everything landed (MemPalace #599)
- ingested pages and PDFs enter recall with no prompt-injection guidance (Supermemory, MemPalace)

## Around it

Before: document-extraction.pdf-to-text, storage.upload. After: agent-memory.user-facts-recall.

## Where this comes from

Weight 8 of 100 in Agent memory: 4 of 36 accounts (MemPalace #599, Graphiti #290, Cognee #633); most listings claim it but few users wrote about it

As data: https://www.anchorterminal.com/api/v1/use-cases/agent-memory.ingest-records.json, or the MCP tool get_use_case with id agent-memory.ingest-records. 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
