mem0
In its own words: “Enhances AI assistants and agents with an intelligent memory layer, enabling personalized AI interactions.” Add extracts memories from conversations; search returns the memory text, a score, created_at and categories.
mem0 adds a memory layer to assistants so they remember preferences and adapt over time. It ships as a library, a self-hosted server and a hosted platform, with SDKs, a plugin ecosystem and published benchmarks. Okto Neuron does a narrower job: turn a Markdown vault into a graph and answer from it with a pointer to the exact bytes. This page marks where the two overlap and where they do not compete at all.
mem0 extracts and stores memories about the people using an app, and ships SDKs, a hosted platform and a self-hosted option. Okto Neuron builds a graph from your own Markdown notes on your machine and returns a pointer to the source bytes with each answer.
Pick mem0 if you are building an app that should remember each user across sessions, or you want a managed service.
Pick Okto Neuron if the memory is your notes and you need to check where every answer came from, locally.
mem0's unit is a memory: a string the LLM extracted from a conversation, stored with a score, a timestamp and categories. Okto Neuron's unit is a claim anchored to a block in a file you wrote, with a byte range and a hash. One is built to learn about a person; the other is built to be checked.
In its own words: “Enhances AI assistants and agents with an intelligent memory layer, enabling personalized AI interactions.” Add extracts memories from conversations; search returns the memory text, a score, created_at and categories.
The graph is derived from your Markdown vault and can be rebuilt from it. ask returns text with a synthesis_status, and with include_sources=true, the vault-relative path, byte range and content hash of every source block.
Competitor cells paraphrase the mem0 README, docs and pricing page. Okto Neuron cells were checked against the 0.3.0 source on 2026-09-26 (storage backends, supersession fields, loopback runtime probing, telemetry extra, MCP surface).
| Dimension | mem0 | Okto Neuron |
|---|---|---|
| Where data lives | Three paths: the open-source library on your machine, a self-hosted server via Docker Compose, or the Mem0 cloud platform. The MCP path is hosted only: nothing runs on your machine and memories live in your Mem0 account | Markdown vault on your disk; the graph is derived from it. Nothing is hosted |
| Storage | Open source: local Qdrant at /tmp/qdrant and a SQLite history store at ~/.mem0/history.db. Server: Postgres with pgvector. Graph memory removed from open source, no queryable graph | Okto Grafx by default (embedded). Ladybug selectable. Neo4j selectable with the neo4j extra, which the installer does not add. The graph is derived and can be regenerated from the files |
| Agent access | Python and JavaScript SDKs, REST. MCP through the hosted server over HTTPS; clients listed: Claude, Claude Code, Codex, Cursor, Windsurf, VS Code, OpenCode. The Claude Code plugin needs a Platform API key | MCP with five tools over a bearer-authenticated port; REST and UI on loopback. The installer wires Claude Code |
| Provenance unit | An extracted memory string with memory, score, created_at and categories, plus a per-memory change history. No pointer back to source text found | A block: vault-relative path, byte_start, byte_end and a sha256 content_hash, returned with include_sources=true on ask |
| Temporal model | created_at, updated_at, expiration_date. Open source is single-pass ADD-only: one LLM call, no update or delete, memories accumulate. Temporal reasoning and Dream (supersede and merge) are platform-only | Ingest-time supersession bounded to the document being edited: valid_until on the old claim, a supersedes edge from the new one, after an LLM correction judge confirms the change. The correction judge does not sweep the whole vault |
| Extraction | LLM, default OpenAI gpt-5-mini with text-embedding-3-small. Ollama supported locally | LLM curation at ingest against an OpenAI-compatible endpoint, loopback runtimes probed first. Synthesis on ask reports synthesis_status |
| License | Apache 2.0 | Elastic License 2.0 from 0.3.0 (source-available). Releases up to 0.2.0 stay Apache 2.0 |
| Hosted option | Hobby free (10,000 adds and 1,000 retrievals per month), Starter at $19/month, Pro at $249/month, Enterprise custom | None |
| Telemetry | MEM0_TELEMETRY, on by default, set to false to turn off | No analytics or telemetry dependency in the shipped package |
| Benchmarks | Publishes LoCoMo and LongMemEval results; its README says the scores reflect its managed platform (README) | LoCoMo categories 1-4 from our own harness and judge, on the benchmarks page. Left out of this table on purpose: a different judge, scope and harness mean the numbers do not compare with self-reported figures |
| Maturity | Established product with SDKs, a hosted platform and a client plugin ecosystem | Prerelease 0.3.0. The upgrade from 0.2.0 has been rehearsed on Linux with a locally built wheel; macOS and Windows are not yet verified on 0.3.0 |
mem0 ships all of these. Okto Neuron has not started most of them.
Hobby is free, and the platform adds temporal reasoning and Dream, which supersedes outdated facts and merges duplicates. Okto Neuron has no hosted option. It runs only on your machine and needs no account.
Python and JavaScript SDKs, REST, and a hosted MCP that lists Claude, Claude Code, Codex, Cursor, Windsurf, VS Code and OpenCode. Okto Neuron's installer wires Claude Code; other clients are unverified.
mem0 publishes LoCoMo and LongMemEval scores. They are footnoted as platform figures, which is a fair caveat. Okto Neuron publishes LoCoMo only, from its own harness and judge, and nothing on LongMemEval. The two sets of numbers were produced differently, so this page does not put them side by side.
A released product with a plugin ecosystem, a CLI and an enterprise tier. Okto Neuron is a 0.3.0 prerelease with five MCP tools.
mem0's search returns the extracted memory string and its metadata; no pointer to the source text was found in the docs reviewed. Okto Neuron returns the vault-relative path, byte_start, byte_end and a sha256 content_hash for each source block.
mem0 open source is single-pass ADD-only; superseding outdated facts is a platform feature. Okto Neuron supersedes on ingest, locally, inside the document being edited, once an LLM correction judge confirms the change, with valid_until on the old claim and a supersedes edge. Cross-document corrections are deferred, not applied.
mem0's MCP path is hosted by Mem0 and memories live in your account. Okto Neuron's MCP server runs on your machine on a bearer-authenticated port, with REST and UI on loopback. The library and server paths of mem0 are self-hostable; only its MCP path is not.
Every ask stamps retrieval.synthesis_status: ok, empty, provider_error, truncated, abnormal_stop or no_llm. An unreachable model returns empty text and a reason, never a paragraph that looks finished.
You want an assistant to learn a user's preferences from conversations and adapt.
You want SDKs in Python or JavaScript, REST, or a hosted MCP that many clients can reach.
You want a free hosted tier to start with and a platform that supersedes and merges memories for you.
You need LongMemEval results, or figures from a managed platform.
Your knowledge is a folder of Markdown and each cited source has to come back with its file path and byte range.
You want supersession to happen locally, on ingest, without a hosted tier.
You want the MCP server and the graph on your machine, you choose where LLM calls go, and you are fine with a prerelease.
You want a failed synthesis to say so in the payload.
No. Our LoCoMo results come from our own harness and a local judge, on categories 1-4 only. Figures other projects publish use their own setup. Ours are on the benchmarks page.
It is source-available. From 0.3.0 the license is the Elastic License 2.0 with an addendum covering SaaS, competing services, internal use and branding. Releases up to 0.2.0 stay under Apache 2.0. The source is on GitHub.
No. It runs on your machine, binds 127.0.0.1 and needs no account.
Every competitor statement on this page comes from one of these pages, read on 2026-09-19, with key rows re-checked on 2026-09-26. If a source has changed since, the page is wrong and we would like to know.
| Claim area | Primary source |
|---|---|
| Positioning, three deployment paths, benchmarks and footnote | https://raw.githubusercontent.com/mem0ai/mem0/main/README.md |
| Open-source overview, local stores | https://docs.mem0.ai/open-source/overview |
| Platform vs open source, graph memory, ADD-only extraction | https://docs.mem0.ai/platform/platform-vs-oss |
| Hosted MCP | https://docs.mem0.ai/platform/mem0-mcp |
| Add operation | https://docs.mem0.ai/core-concepts/memory-operations/add |
| Search operation and returned fields | https://docs.mem0.ai/core-concepts/memory-operations/search |
| Search API reference | https://docs.mem0.ai/api-reference/memory/search-memories |
| Temporal reasoning, platform-only | https://docs.mem0.ai/platform/features/temporal-reasoning |
| Ollama support | https://docs.mem0.ai/components/llms/models/ollama |
| Claude Code integration | https://docs.mem0.ai/integrations/claude-code |
| CLI and MEM0_TELEMETRY | https://docs.mem0.ai/platform/cli |
| Pricing | https://mem0.ai/pricing |
Prerelease 0.3.0. Free to run locally. No account required.