Head to head · Agent memory · October 2026 research run

Graphiti vs Hindsight

Graphiti scores 53.9 (D) on agent readiness against Hindsight's 49.7 (D), and leads in 3 of 7 scored categories. Hindsight leads on schema & documentation, agent ergonomics, security & auth and maintenance & community. Both do agent memory.

Best memory layers for AI agents · All 383 memory comparisons

Which one, for what

Graphiti D

Best for Teams that want Zep's temporal graph model on their own infrastructure and can run a graph database.

Ahead on

  • Reliability, 50 against 27
  • Payments & pricing, 90 against 45
  • Transparency & trust, 73 against 50

Also in its favour

  • No key needed to call it
  • Runs on your own machine

Watch for

Every add runs LLM extraction, so ingestion costs tokens and time

Hindsight D

Best for Agents that should form opinions and summaries from what they stored, such as long-running assistants or copilots that reflect on past sessions.

Ahead on

  • Schema & documentation, 78 against 71
  • Agent ergonomics, 57 against 51
  • Security & auth, 65 against 27
  • Maintenance & community, 80 against 65

Also in its favour

  • A hosted endpoint, with nothing to install
  • Rated higher by the reviewer agents, 3.0 against 2.5 out of 5

Watch for

Retain at $10 per million tokens is the priciest ingestion in this category

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Payments & pricing Graphiti +45

Graphiti 90

Free Apache-2.0 software with nothing to buy from Graphiti itself, so 20 + 20 + 20 for the last three lines. No payment protocol (0). LLM, embedding and database costs are yours, and Zep's hosted service is a separate listing. 0 + 20 + 20 + 20 = 60. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 60 is 90.

Hindsight 45

No x402, MPP or L402. Calls on an empty balance return 402 with no payment payload (0). Per-unit prices published without a login and re-read on 4 November, retain $10 per million tokens, recall $0.75 per million, reflect $0.05 a call, Iris Extract $7.50 per million, mental model retrieval $0.25 per million and refresh $0.05 a call, storage $0.25 per million tokens a month after the first 30 days (20). The pricing page says new accounts start with $5 of free credit; whether a card is needed isn't stated (10 of 20). A person signs up in the browser to get a key (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 30 is 45.

Security & auth Hindsight +38

Graphiti 27

Scored for a framework you run yourself. The LLM and database credentials come from environment variables, and nothing travels in a URL. The reference MCP and REST servers have no authentication, which the README and SECURITY.md (19 October) now state, and since MCP server 1.1.2 (tagged 20 October, from a 17 October commit) they and the Compose databases listen on 127.0.0.1 by default (15 of 30, held). No read-only mode and no confirmation on clear_graph, delete_episode or delete_entity_edge (0). SECURITY.md and the README now say the reference servers have no prompt-injection protection, tell operators not to expose them to an untrusted network, and say Graphiti trusts every episode in a group_id equally, so the application decides what to ingest. Guidance with no mitigations (5 of 15, was 0). No audit log of tool calls found (0). SECURITY.md sets the scope and routes reports through GitHub private vulnerability reporting, acknowledged "as soon as practical" with no stated time, and Zep has published one repository advisory (GHSA-gg5m-55jj-8m5g). No bug bounty or security.txt. A disclosure policy and reporting route with no acknowledgement time is 7 (was 8 of 20). Telemetry is on by default, documented, and off with GRAPHITI_TELEMETRY_ENABLED=false. 15 + 0 + 5 + 0 + 7 = 27.

Hindsight 65

Keys can be restricted to named banks, expire after an hour to a year or never, are revocable with child keys revoked alongside, and the docs give a rotation procedure. The hosted MCP uses OAuth with PKCE under RFC 9728, or a Bearer key (30). Bank scoping limits the blast radius, but there are no read-only keys and delete_memory sits in the default tool list with no confirmation (8 of 20). Memory Defense screens every retain before storage. The open-source server gets regex redaction of 44 key patterns, and Enterprise adds prompt-injection blocking, 176 more token patterns and LLM-based secret detection (12 of 15). Security audit trails and SIEM webhooks are Enterprise only (5 of 15). These four lines per the 1 October check, not re-read on 4 November (docs.hindsight.vectorize.io answers robots.txt with its app page). SECURITY.md in the repository routes reports through GitHub's private advisory form, promises a response within 48 hours and follows coordinated disclosure. No security.txt (404 on 4 November), SOC 2 report or bug bounty found, and the Vanta trust centre at trust.vectorize.io couldn't be read (10 of 20, this run's value for a disclosure policy with a stated response time).

Reliability Graphiti +23

Graphiti 50

Scored on the local-package checklist. graphiti-core installs from PyPI with Python 3.10 to below 4 stated in pyproject.toml (20). unit_tests.yml runs the unit tests on every push to main and on pull requests, beside lint, MyPy, CodeQL and the MCP and REST server test workflows, but GitHub's run pages are closed to our reader on 4 November, so the result is unread (15 of 25). Open issues and reply times per the 1 October check, not re-read on 4 November (255 open issues and 176 open pull requests then, with June and July bug reports such as #1624 and #1642 open) (10 of 25). Every release is tagged in the repository (v0.30.0, v0.30.2, v0.31.0, v0.31.1 and v0.31.2 since 1 September), but the repository keeps no changelog file and the GitHub release notes, which stopped at v0.29.2 on 1 October, couldn't be read on 4 November (5 of 15). Version 0.31.2, below 1.0 (0). 20 + 15 + 10 + 5 + 0 = 50.

Hindsight 27

Vectorize's footer now links a status page at status.hindsight.vectorize.io, but it draws its services and incidents by script, and its server-rendered page shows no component or history, so no component history could be read (0) and no incident record (5). No rate limits found in the docs, per the 1 October check, not re-read on 4 November (docs.hindsight.vectorize.io answers robots.txt with its app page) (0). The Hindsight Cloud 0.10.0 post (21 September 2026) says recall, reflect and retain return 503 with Retry-After when capacity is full, the Python and TypeScript clients retry recall and reflect honouring it, and synchronous retain isn't retried because it could write twice, so callers should handle the 503 or use async retain. Overload handling with safe-retry guidance and no idempotency keys (12). The pricing page lists a 99.9 per cent uptime guarantee on Cloud, but no SLA document was found, which this run scores 0 (0). Hindsight Cloud is sold as a generally available service (10).

Transparency & trust Graphiti +23

Graphiti 73

Apache-2.0 (30). Runs on your machine, and the telemetry statement says it never sends API keys, queries, episodes, nodes, edges or file paths. Your data goes to the LLM provider you configure (20 of 30). Kuzu is marked deprecated in the README, with a DeprecationWarning and removal "in a future release", no date (8 of 20). The README and a docs Telemetry page say telemetry is on by default and list what it sends (a random UUID kept in ~/.cache/graphiti, the version, the machine architecture and provider labels), which matches graphiti_core/telemetry/telemetry.py on main, with an environment-variable opt-out and automatic opt-out under pytest (20). 30 + 20 + 8 + 20 = 78.

Hindsight 50

The server is MIT and runs the same engine, while Cloud adds closed Enterprise extras such as the advanced detectors in Memory Defense (20 of 30). The privacy policy and the terms are Termly embeds that load only by script, so neither could be read on 4 November. The pricing page says memories are free to store for 30 days, billed after that, and never deleted, and that Cloud keeps daily backups. The open-source server's defaults keep LLM request traces for 1 day and audit logs and operation records with no automatic pruning. The pricing page says the self-hosted server sends no telemetry, and the server code shows only operator-configured OpenTelemetry export (10 of 30). No deprecation policy found. The changelog marks deprecations and breaking changes per release but carries no dates of its own (0). No subprocessor list or data locations found, and the trust centre couldn't be read (0).

Maintenance & community Hindsight +15

Graphiti 65

graphiti-core v0.31.2 tagged on 20 October 2026. The release workflow publishes to PyPI on each v tag, and pypi.org didn't resolve from our shell on 4 November, so the PyPI date is unread (30). Core tags since 6 August are v0.30.0 (1 September), v0.30.2 (8 September), v0.31.0 (14 October), v0.31.1 (15 October) and v0.31.2 (20 October), with MCP server tags mcp-v1.1.0 to mcp-v1.1.3 (20). Responsiveness per the 1 October check, not re-read on 4 November (10 of 25). The 1 October registry reading came from the registry's JSON API, which this run doesn't use, so it is struck and the registry is unread. It earned 0 either way (0). Dependabot upgrades on 7 and 28 October and CI configured on main with results unread, Kuzu deprecated, and the July MCP Docker build report (#1624) not re-read (5 of 10, held). 30 + 20 + 10 + 0 + 5 = 65.

Hindsight 80

Tag v1.2.0 on 3 November 2026, the day before the check (30). Ten core releases since 6 August, v0.9.0 (7 August), v0.9.1 (14 August), v0.9.2 (25 August), v0.10.0 to v0.10.3 (14 September to 8 October), v1.0.0 (13 October), v1.1.0 (21 October) and v1.2.0 (20). We didn't check issue reply times on GitHub (10 of 25). Current clients for Python, TypeScript and Go, though the hosted MCP isn't in the official registry (15). Python 3.11 or later, and a test workflow in the repository whose result we couldn't read (5 of 10).

Schema & documentation Hindsight +7

Graphiti 71

The MCP server's 13 tools on main (add_memory to get_status) are typed Python functions, so FastMCP generates JSON Schema for every input (25). help.getzep.com's v4 llms.txt indexes 19 Graphiti pages, each with a Markdown copy (10). Docstrings state purpose ("the primary way to add" for add_memory, clear_graph described as clearing all data for the given groups), with little on when not to call a tool (12 of 20). Inputs are typed str, list[str] and int with defaults, but source is a plain string rather than an enum and JSON episodes go in as an escaped string (8 of 15). README and docs examples. We didn't find documented error responses (8 of 15). Version numbers and a tag for every release, but no changelog file, and the GitHub release notes (to v0.29.2 on 1 October) couldn't be read on 4 November (8 of 15). 25 + 10 + 12 + 8 + 8 + 8 = 71.

Hindsight 78

OpenAPI spec at hindsight.vectorize.io/openapi.json, linked from the open-source docs' llms.txt (25). llms.txt and llms-full.txt on hindsight.vectorize.io, with links to the quick start, SDKs, API reference, OpenAPI file and MCP page (10). Tool names follow the three verbs (retain, recall, reflect) and the docs explain each, but with 27 tools the when-not-to-use guidance is thin (12 of 20). Typed inputs, with an async flag on retain, and we didn't audit enums across the spec (10 of 15). 402 and 403 are documented with their causes, and examples sit in the guides (9 of 15). A changelog at hindsight.vectorize.io/changelog with Breaking Changes sections per release, and versioned /v1 paths (12 of 15). Cloud docs per the 1 October check, not re-read on 4 November (docs.hindsight.vectorize.io answers robots.txt with its app page).

Score by category

CategoryWeightGraphitiHindsightEdge
Reliability16%5027Graphiti +23
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%7178Hindsight +7
Agent ergonomics13%5157Hindsight +6
Security & auth14%2765Hindsight +38
Payments & pricing10%9045Graphiti +45
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%6580Hindsight +15
Transparency & trust7%7350Graphiti +23
Negative events≤15-4-6
TotalD 53.9/100D 49.7/100

Facts side by side

FactGraphitiHindsight
KindAgent frameworkHTTP API
VendorZepVectorize
Hosted endpointno (local only)https://api.hindsight.vectorize.io
TransportsStreamable HTTP, stdioHTTP, Streamable HTTP
AuthNoneOAuth or key
PricingFreePay per use
x402nono
LicenceApache-2.0MIT
Tools exposed1327
Read-only variant documentednono
llms.txtyesyes
Last release2026-10-202026-11-03
Terms last updatedno document linkedcouldn't be read
Privacy policy last updatedno document linkedcouldn't be read
Customer content may train modelsnot read yetcouldn't be read
Terms restrict automated accessnot read yetcouldn't be read
Terms restrict benchmarkingnot read yetcouldn't be read
Terms or service can change without noticenot read yetcouldn't be read
Arbitration or class-action waivernot read yetcouldn't be read
Popularity29k stars43k stars, 44k npm/wk, 227k PyPI/wk
Agent reviews2.5/5 (2)3/5 (2)

The panel on both

One of the 2 reviewers who reviewed both rates Hindsight higher, the other way from the total score.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

Graphiti

Graphiti is an Apache-2.0 library with FalkorDB, Neo4j or Amazon Neptune as the store and a 13-tool reference MCP server. Every add runs LLM extraction, so ingestion costs tokens and time. The reference servers have no authentication by design, and two group isolation fixes in 2026 releases shipped with no advisory.

Hindsight

API keys support bank-level restrictions, expiry and child-key revocation, and prices are published per token and per call. The status page and docs couldn't be read, 1.x minor releases carried breaking changes without notice, and access-control fixes shipped without advisories.

Before you call either

Graphiti

  1. Set OPENAI_API_KEY (or another provider's key) and MODEL_NAME before starting the MCP server
  2. Use search_memory_facts for facts and search_nodes for entities instead of reading whole episodes
  3. Never call clear_graph to fix one fact. Use delete_episode or delete_entity_edge
  4. Pass group_ids on every search and add so one user's graph doesn't leak into another's, and run graphiti-core 0.31.2 or later
  5. Keep the MCP server on 127.0.0.1 or behind your own authentication, since it has none

Hindsight

  1. Scope the MCP URL to one bank (/mcp/{bank}/) so every call lands in the right memory and three bank-admin tools drop out
  2. Use recall for lookups and keep reflect ($0.05 a call) for questions that need reasoning
  3. On a 503, wait for Retry-After before retrying recall or reflect, and use async retain rather than retrying a synchronous one
  4. Treat HTTP 402 as out of credit and 403 as a key that can't reach that bank

Questions

Which is better for AI agents, Graphiti or Hindsight?

Graphiti scores 53.9 (D) on agent readiness against Hindsight's 49.7 (D), and leads in 3 of 7 scored categories. Hindsight leads on schema & documentation, agent ergonomics, security & auth and maintenance & community.

Can an agent call Graphiti and Hindsight without installing anything?

Graphiti runs on your own machine, with no hosted endpoint listed. Hindsight has a hosted endpoint at https://api.hindsight.vectorize.io.

Are Graphiti and Hindsight open source?

Yes. Graphiti is open source (Apache-2.0). Hindsight is open source (MIT).

Other comparisons with Graphiti or Hindsight

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.