Mengram
Mengram is a self-hostable AI memory API that gives Claude, Cursor, Codex and ChatGPT one shared memory graph.
Procedural memory is where Mengram pulls ahead of most memory layers — failures actually rewrite the stored workflow instead of sitting in a log. The September 2026 execution-policy work adds per-version trust labels, a policy gate and a regression gate on top. Free tier is testable with real work, and $5/mo Starter or $19/mo Pro is cheap for memory shared across four different tools. Skip it if you need SOC 2 or HIPAA paperwork today.
Verified 4d ago · liveness 75/100 · cite: rightaichoice.com/tools/mengram
- Developers who want Claude, Cursor, Codex and ChatGPT reading from one persistent memory graph
- Claude Code users who keep losing context to /clear and auto-compaction
- Teams building multi-agent systems on LangChain or CrewAI that need isolated memory per end-user
- SaaS founders replacing a chunking-and-embedding RAG pipeline with a REST or MCP memory backend
- Organizations that need SOC 2 or HIPAA certification out of the box
- Non-technical users looking for a no-code memory product
- High-volume production agents on Free, Starter or Pro — 1,000 memory adds/month at Pro runs out fast
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Mengram if you need SOC 2 or HIPAA certification on day one, or if your agent writes a lot of memory and you don't want to self-host past roughly 1,000 adds a month.
Memory adds are the real meter, not searches: Free gives 40/month and Pro 1,000/month, so a busy Claude Code workflow can exhaust the tier you paid for
Free at $0/mo is a genuine test tier with unlimited isolated user_ids, Starter at $5/mo fits a solo developer, and Pro at $19/mo is where production work starts because procedure evolution and smart triggers live there. That undercuts hosted memory layers priced per seat at the high end, but Growth at $59/mo and Business at $99/mo are where volume pricing resembles an infrastructure bill rather than a dev tool. Annual billing takes 20% off the monthly rates.
In short
Mengram — Mengram is a self-hostable AI memory API that gives Claude, Cursor, Codex and ChatGPT one shared memory graph. Best for Developers who want Claude, Cursor, Codex and ChatGPT reading from one persistent memory graph, Claude Code users who keep losing context to /clear and auto-compaction, Teams building multi-agent systems on LangChain or CrewAI that need isolated memory per end-user. Free to start; paid plans from $5/mo.
What's new in Mengram
Checked 2 days agoAcross the latest 1 update: 1 feature update.
What people actually say about Mengram — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
34 mentions across 5 sources (Hacker News, YouTube, Product Hunt, GitHub, Lemmy) · researched Jul 6, 2026.
Average across the 5 sources that answered — each source counts once, not each post.
- +Three memory types (semantic, episodic, procedural) are more human-like than flat vectors.
- +Cognitive Profile generation makes personalization easy with one API call.
- +Low latency (<50ms) for memory operations, as claimed in documentation.
- +Multilingual support with 23 languages via Cohere embeddings is a standout feature.
- +MCP server with 30 tools integrates directly into Claude Code and Cursor.
- −MCP server returns Cloudflare 403 error for some users, breaking core functionality.
- −Self-hosting documentation is unclear; version mismatches reported.
- −Hard edge cases like contradictions and staleness lack built-in resolution.
- −Very small community (183 stars, 20 issues) means limited support resources.
- −LangChain integration example code was requested but not immediately available.
- • Free tier is limited to 40 memory adds/month — likely too small for serious use.
- • Self-hosting requires own infrastructure; official docs may lack clarity on API key usage.
Viability Score
How well maintained and how widely used is Mengram? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- One-click memory connector for claude.ai and Claude Desktop — no terminal, no API key pasted
- Claude Code plugin with SessionStart and Stop hooks that survive /clear and auto-compaction
- MCP server exposing 30 memory tools for Claude Desktop, Cursor, Codex and Windsurf
- Three memory types: semantic (facts), episodic (events) and procedural (workflows)
- Procedural memory that auto-evolves workflow versions after failures, with version history and evolution log
- Cognitive Profile generation: one API call returns a ready-to-use system prompt from all three memory types
- Execution policy with per-version trust labels, a policy gate and a regression gate
- Knowledge graph storing entities and relations, not just text chunks
- Smart Triggers for reminders, contradiction alerts and workflow pattern detection
- Autonomous memory agents: Curator (cleans contradictions), Connector (finds patterns), Digest (weekly briefs)
- Multilingual retrieval across 23 languages via Cohere multilingual embeddings, including cross-lingual search
- Import ChatGPT exports, Obsidian vaults and text files by CLI
- Local mode: one command reads existing Claude Code session history with no account, key or upload
- REST API, Python SDK and JavaScript SDK with webhooks
- Per-user memory isolation via user_id for multi-tenant agent products
About Mengram
Mengram is an AI memory API for developers who are tired of re-explaining their stack, constraints and preferences to every new AI session. It connects to the assistants you already use through MCP or a REST API, extracts what matters, and writes it to one shared graph instead of leaving each tool with its own private, quickly forgotten context. Claude, Cursor, Codex and ChatGPT all read and write to that same store. The graph is organized into three memory types: semantic (who you are, your preferences, your stack), episodic (what happened — events, decisions, outcomes) and procedural (how you work, stored as workflows that evolve when a run fails). A single API call returns a Cognitive Profile, a ready-to-use system prompt assembled from all three. Other pieces worth knowing: a knowledge graph of entities and relations rather than loose text, Smart Triggers for reminders and contradiction alerts, and autonomous Curator, Connector and Digest agents that clean, connect and summarize. Retrieval works across 23 languages via Cohere multilingual embeddings. Setup splits by tool. Claude users get a one-click connector on claude.ai and Claude Desktop — paste the URL, leave the OAuth fields empty, sign in with an email code, no terminal. Claude Code, Cursor and Windsurf install via a plugin or CLI, and the Claude Code plugin wires SessionStart and Stop hooks so context returns after /clear and auto-compaction. For teams shipping agent products, you pass a user_id and each end-user gets isolated facts, events, workflows and their own cognitive profile — server-side extraction means no chunking and no embedding pipeline. It is Apache-2.0 and self-hostable, so it sits next to Mem0, Zep and Letta rather than the closed memory features baked into individual assistants. The trade-off is the usual open-source one: flexibility and no lock-in in exchange for running and tuning the layer yourself if you outgrow the hosted tiers.
Behind the Verdict
The pitch that lands first is the shared graph. If you rotate between Claude, Cursor and Codex all week, the pain of re-stating your stack every session is real, and Mengram is aimed straight at it. The one-click Claude connector is the low-friction path, and the Claude Code plugin's SessionStart hook is the part that matters most in daily use — context coming back after a compact is the difference between a memory layer you keep and one you uninstall. Where it earns its keep is the procedural side. Storing workflows and revising them on failure is a genuine step past flat fact storage, and the September 2026 execution-policy additions — trust labels per version, a policy gate, a regression gate — are what you want before letting an evolved procedure run unsupervised. Pick Mengram if you are building agent products and need per-user isolation: pass a user_id, get isolated facts, events and workflows, no chunking pipeline to maintain. In practice, budget for the limits. Pro gives you 1,000 memory adds and 10,000 searches a month; high-volume production agents on the lower tiers will burn through adds quickly, and reranking only arrives at Pro via LLM reranking or Growth via Cohere cross-encoder. If your agent writes memory on every turn, watch the add count, not the search count. Self-hosting is the honest answer to compliance questions. Apache-2.0 means no vendor lock-in, but running it is on you — it is a workaround, not a certification. Teams that need signed paperwork on day one should look at vendors that ship it, or self-host behind their own controls. Compared with Mem0, Zep and Letta, the differentiators worth testing are procedural evolution, the autonomous Curator/Connector/Digest agents, and the genuinely broad tool support. Mem0 is the more common
Researching Mengram? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Mengram actually fits — and what changes day-one when you adopt it.
Installs the Claude Code plugin, points it at a free API key, and runs the CLI import against existing Claude Code session history so the memory graph already knows the projects and stack.
Outcome: After /clear or an auto-compaction, the SessionStart hook re-injects context and the next session starts with the stack and workflow patterns instead of from zero.
Calls the REST API with a distinct user_id per end-user, relying on server-side extraction so there is no chunking or embedding pipeline to maintain, then reads back a Cognitive Profile for the system prompt.
Outcome: Each customer's agent recalls that customer's own facts, events and workflows without cross-user leakage, shipped in roughly ten lines rather than a LangChain RAG stack.
Wires Mengram in as the shared memory backend, uses Smart Triggers for contradiction alerts, and applies the execution-policy gate so low-trust workflow versions cannot run unchecked.
Outcome: Crew members share decisions and outcomes, and a regression gate catches one workflow revision breaking another before it reaches production.
Use Cases
- Give your AI assistant persistent memory of preferences, stack and project context across sessions and machines
- Keep Claude, Cursor and Codex reading from one memory graph instead of three separate silos
- Add per-user memory scoping to a SaaS agent so every end-user has isolated facts and workflows
- Build multi-agent crews on LangChain or CrewAI that recall past decisions automatically
- Replace a chunking-and-embedding RAG pipeline with a memory API plus a Cognitive Profile call
- Gate agent workflow versions so a revision that regressed previously working runs is blocked before it executes
- Import existing Claude Code history so memory knows your projects from the first session
- Build voice assistants on Vapi that remember user facts and conversation history across calls
Models Under the Hood
as of 2026-10-17
Limitations
- Memory writes are metered and the ceilings are low on the entry tiers: 40 adds/month on Free, 100 on Starter, 1,000 on Pro, 3,000 on Growth, 8,000 on Business.
- Procedure evolution and smart triggers only unlock at Pro ($19/mo) and above, so the procedural-memory capability that most distinguishes Mengram cannot be evaluated on Free or Starter.
- Retrieval quality is tiered too — Free and Starter run vector search with no reranking, Pro adds LLM-powered reranking, and only Growth and Business use Cohere cross-encoder reranking.
- Rate limits step from 20 to 300 req/min across the tiers.
- SSO and access controls appear only on Enterprise, alongside data residency and on-premise deployment.
as of 2026-10-07
Verification history
We have re-verified Mengram 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Mengram tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free
$0/mo
Ideal for
Developers evaluating Mengram with real work, and anyone who wants to test per-user memory isolation before paying
What this tier adds
Starting tier: 40 memory adds and 200 searches a month, 3 agent runs, unlimited isolated end-users, 20 req/min, no reranking and no procedure evolution
Starter
$5/mo
Ideal for
A solo developer or indie project running one or two AI tools with light memory volume
What this tier adds
Adds volume to Free — 100 adds and 500 searches a month, 10 agent runs, 60 req/min, plus 2 webhooks and 1 team
Pro
$19/mo
Ideal for
Production apps and power users who need procedural memory and triggers to actually function
What this tier adds
Unlocks procedure evolution and smart triggers, adds LLM-powered reranking, and raises limits to 1,000 adds, 10,000 searches and 120 req/min
Growth
$59/mo
Ideal for
Scaling apps that have outgrown 1,000 memory adds a month and want better retrieval ranking
What this tier adds
Switches reranking to Cohere cross-encoder, makes agent runs unlimited, and raises limits to 3,000 adds, 20,000 searches, 200 req/min, 25 webhooks and 10 teams
Business
$99/mo
Ideal for
Teams and high-volume applications that need headroom plus more webhooks and unlimited teams
What this tier adds
Raises volume to 8,000 adds, 30,000 searches and 300 req/min, and lifts webhooks to 50 with unlimited teams
Enterprise
Custom
Ideal for
Organizations with compliance, residency or infrastructure requirements that rule out the shared tiers
What this tier adds
Custom limits and rate limits, dedicated infrastructure, SSO and access controls, priority support with SLA, custom integrations, data residency and on-premise deployment
Where the pricing makes sense
The company stage and team size where Mengram's pricing actually pencils out — and where peers do it cheaper.
Free at $0/mo is a genuine test tier with unlimited isolated user_ids, Starter at $5/mo fits a solo developer, and Pro at $19/mo is where production work starts because procedure evolution and smart triggers live there. That undercuts hosted memory layers priced per seat at the high end, but Growth at $59/mo and Business at $99/mo are where volume pricing resembles an infrastructure bill rather than a dev tool. Annual billing takes 20% off the monthly rates.
Setup time & first value
How long it actually takes to get something useful out of Mengram — broken out by persona, not the marketing-page minute.
Claude users get first value in about two minutes: paste the connector URL in Settings → Connectors, leave OAuth blank, sign in with an email code — no terminal and no key. Claude Code users need two commands plus a config file, and the CLI import of existing session history means memory knows your projects from minute one rather than week two. REST and SDK integrations are a short
Switching to or from Mengram
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From ChatGPT built-in memory: import your ChatGPT export via the CLI so past facts and preferences land in the shared graph
- →From Obsidian vaults: run the one-command vault import to seed semantic memory from your existing notes
- →From a hand-rolled RAG pipeline: replace chunking and embedding with a server-side extraction call plus a Cognitive Profile fetch
- →From .cursorrules files: keep the rules file for conventions and add Mengram as the dynamic layer that remembers session history
- →From Claude Code session history: pip install mengram-ai and run the import command to backfill projects and workflows
- ↗To Mem0: export memory through the API and reformat into Mem0's add/search calls, accepting the loss of procedure evolution
- ↗To Zep: map semantic and episodic entries onto Zep's session model and re-create workflows manually, since procedural evolution has no direct equivalent
- ↗To Letta: rebuild memory blocks from a Cognitive Profile export and reimplement triggers yourself
- ↗To self-hosted Mengram: run the Apache-2.0 deployment on your own infrastructure and keep the same API surface
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Mengram”, and we withheld 6: 6 could not be judged, because “Mengram” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Mengram.
Official links
Tools that pair well with Mengram
Common stack mates teams adopt alongside Mengram, with the specific reason each pairing earns its keep.
Agentmemory
Agentmemory gives coding agents persistent memory between sessions — local, open-source recall for Claude Code, Cursor and Codex.
MemoryLake
MemoryLake is portable AI memory for ChatGPT, Claude, Codex and Gemini — encrypted, exportable, and shared across every agent you run.
Signetai
Local-first agent memory and secrets layer that gives Claude Code, Codex, and OpenCode one shared context on your own machine.
Featured Head-to-Head Comparisons
Mengram vs Spider Cloud
Mengram and Spider Cloud serve fundamentally different needs: one provides persistent memory for AI agents, the other live web data. If your priority is giving agents long-term recall across sessions, start with Mengram's free tier and scale up. If you need real-time web scraping with structured output for RAG, Spider Cloud is the clear choice thanks to its cheap per-page cost and extensive catalog. For most teams, you'll likely need both eventually.
Mengram vs Presto Voice
These tools serve completely different needs. Choose Presto Voice if you run a QSR chain and want proven drive-thru automation with upselling. Choose Mengram if you're a developer seeking a free, open-source memory layer to make AI agents persistent across sessions.
Mengram vs Temporal Ai
Choose Temporal AI if you need a durable execution engine to build crash-resistant workflows and AI agents with automatic retries and rollbacks — it's production-proven by OpenAI and Replit. Choose Mengram if your priority is giving AI agents persistent, human-like memory that works across conversations and tools, without setting up RAG pipelines. They solve different problems: Temporal for reliability, Mengram for memory.
Alternatives to Mengram
View allAgentmemory
Agentmemory gives coding agents persistent memory between sessions — local, open-source recall for Claude Code, Cursor and Codex.
MemoryLake
MemoryLake is portable AI memory for ChatGPT, Claude, Codex and Gemini — encrypted, exportable, and shared across every agent you run.
Frequently Asked Questions
Categories
Topics
Used Mengram? Help shape our editorial sentiment research.