Maya prefers Mercury — better 1099 export
Got it. Noted as a banking preference for Maya.
Saved to mnueron
Episodic
Every conversation captured. Your AI saves what was said, when, by whom — searchable forever.
New
Session Continuity: auto-backup your Claude Desktop / Cowork sessions every 4 hours — “Resume from summary” without losing the thread.
How it works →Local-first AI memory · MITOne memory layer.
One memory layer.
Every AI tool.
Your machine.
Persistent memory for Claude Desktop, Claude Code, Cursor, Windsurf, Cline — and any app you build with OpenAI, Anthropic, Mistral, or anything else. Local-first and free forever. Optional hosted backend for cross-machine sync and team sharing.
Already have an account? Open Dashboard to sign in and start viewing your memories, savings, runbooks, and Meet intelligence.
npm install -g mnueron && mnueron setupFull install guide →Three ways to get started
Pick how your AI tools talk to mnueron.
Same memory store, three interfaces. Copy a snippet, run it once, your AI starts remembering.
One command to install, set up your local memory store, and point your AI tools at it.
"hl-c"># Step 1 — install mnueron globally
npm install -g mnueron
"hl-c"># Step 2 — interactive setup (writes ~/.mnueron/config.json)
mnueron setup
"hl-c"># Step 3 — save your first memory
mnueron save "Maya prefers Mercury over Stripe — better 1099 export"
"hl-c"># Step 4 — recall it later from any AI tool
mnueron recall "what does Maya think about banks"Six kinds of memory · one store
Human memory isn't one thing. Neither is ours.
Most memory layers store flat chat logs. mnueron stores six layers, each tuned to a different question your AI will ask later.
At dinner, Maya mentioned she prefers Mercury over Stripe because of the 1099 export.
↓
Extracted fact · preference
Maya prefers Mercury over Stripe (1099 export).
Semantic
Facts extracted from chats. Standalone truths your agent can recall without re-reading the whole conversation.
Maya
Maya P.
maya@stripe.com
Canonical
Maya Patel
3 aliases · 47 mentions
Entities
Every mention of Maya, Maya P., and maya@stripe.com collapses into one canonical person.
Knowledge graph
Relationships between entities. 'Maya works at Stripe' is queryable like a graph, bi-temporal aware.
Push mnueron to Vercel
- 1
cd repo - 2
git add files - 3
git commit -m '...' - 4
git push origin main
Runbooks
Procedural memory. Save a how-to once, your AI agents pull it back when they hear a matching trigger.
Memory 1
Maya prefers Mercury
Memory 2
Maya likes Mercury better
Merged by AI
Maya prefers Mercury bank over Stripe (1099 export).
Self-revising
Duplicates auto-detected. LLM merges them on approval, preserving every distinct fact.
Privacy-first. Powerful by design.
Mnueron is built around you — your data, your workflow, your control.
Local-first by default
Your memories live in ~/.mnueron/memories.db on your machine. No account, no upload, no telemetry. Free forever.
Cross-tool, cross-provider
Save in Claude Code, recall in Cursor. See the same memories on claude.ai via the Chrome extension. Provider-agnostic.
Hybrid semantic search
Local BM25 keyword + sentence-transformer embeddings via sqlite-vec, fused with reciprocal-rank. No OpenAI key needed.
Secret redaction at write time
13 regex patterns + Bearer / Basic-auth handling strip AWS, GitHub, OpenAI, Anthropic, Stripe, Slack, JWT, and PEM keys before they touch storage.
Open source, MIT client
Fork it, embed it, sell what you build with the SDK. The hosted backend is source-available (FSL → Apache 2.0 in 2 years).
Optional hosted upgrade
When you're ready, mnueron migrate-to-hosted uploads your local DB and flips the active provider. Cross-machine sync, no rewrites.
- 100% Local Control
- End-to-End Private
- Works Everywhere
- Built for Developers
Works with
Auto-configured for every AI tool you already have.
Run mnueron setup once. The wizard detects every installed AI tool and wires mnueron into each. Restart your tools and your AI remembers.
Claude Desktop
MCP auto-config
Claude Code
MCP via claude mcp add
Cursor
~/.cursor/mcp.json
Windsurf
~/.codeium/windsurf
Cline
VS Code extension
claude.ai
Chrome extension
chatgpt.com
Chrome extension
Your apps
Python + C# SDK
Chrome extension
Capture from claude.ai and chatgpt.com. Or use them like Claude Desktop already knew you.
The mnueron Chrome extension turns the AI chat sites you already use into first-class mnueron sources. Save what matters, pull in your past conversations, and have the AI quietly know your context as you type.
- 1One-click capture. Click the toolbar icon on any claude.ai or chatgpt.com conversation — captured into your mnueron store, namespaced by site.
- 2Backfill all history. Import every past Claude conversation in one click — pulls via Claude's internal API, ~60 seconds for 100 chats. Idempotent and pausable.
- 3Ambient context. As you type a prompt, mnueron quietly searches your store and surfaces a small pill above the input when relevant memories exist. Click to insert.
- 4Local or cloud. Toggle Local / Hosted in the popup — same UI, just changes where your captures go.
One click from the Chrome Web Store. Source on GitHub — Edge and Firefox builds available there.
mnueronSettings
Backend
mnueron.comconnected
On Claude
Recall
search your memories…
Captures today12
mnueron · memory for every AI
New · Export your memories
Take your memory anywhere.
Export any subset of your memories — by namespace, by conversation, or all of it — to Markdown, HTML, JSON, NDJSON, or CSV. Drop the file into Claude, Cursor, Cline, ChatGPT, a spreadsheet, or your own pipeline. Your memory belongs to you, in whatever format helps you reuse it.
Group by conversation for clean transcripts you can drop into any chat.
Group by namespace when you want one document per project.
HTML → PDF via your browser's print dialog. No extra deps.
Round-trip safe JSON and NDJSON re-import to any mnueron instance.
Export memories
Browser dialogFormat
MD
HTML
JSON
NDJSON
CSV
Group by
Flat
Conversation
Namespace
Namespace
claude-cowork▾
127 memories · grouped into 14 conversations
How it works
Three commands and you're live.
1
Install
npm install -g mnueron && mnueron setup. The wizard configures Claude Code, Cursor, Cline, and any other MCP-aware tool you have.
2
Use AI tools normally
Memory tools appear automatically. Save facts, decisions, project conventions. Memories accumulate at ~/.mnueron/memories.db.
3
(Optional) Go hosted
Want cross-machine sync or team sharing? mnueron migrate-to-hosted uploads your local DB and flips the active provider. One flag.
SDKs
Three lines from your app to provider-agnostic memory.
Drop the Python or C# SDK into any app. The same pattern works with OpenAI, Anthropic, Mistral, Gemini, or anything else — mnueron doesn't know or care which LLM you're using.
from mnueron import Mnueron
from openai import OpenAI
mem = Mnueron(api_key="mnu_...")
llm = OpenAI()
# Pull memory before the LLM call — provider-agnostic
context = mem.search(user_message, namespace=f"user-{uid}", k=5)
ctx_str = "\n".join(m.content for m in context)
resp = llm.chat.completions.create(
model="gpt-4o-mini",
messages=[
{"role": "system",
"content": f"What you know about this user:\n{ctx_str}"},
{"role": "user", "content": user_message},
],
)
# Save anything worth remembering
mem.save(extract_facts(resp), namespace=f"user-{uid}")See it in action
One memory layer. Every AI tool.
On the left: what mnueron looks like across every AI tool you use. On the right: the save → recall → inject loop in action, looping every 12 seconds.
claude.aimnueron connected
What was the auth approach we chose for the side project?
Found 3 memories from 2 weeks ago
from
We chose bcrypt with cost 12 + httpOnly session cookies (24h expiry). Rejected JWT — too many gotchas with refresh rotation for a solo dev.side-projectClaude recalled your past decision without you re-explaining anything.
Memory Consolidation
Memory that maintains itself.
Every chat tool you've ever owned gets worse the longer you use it — history bloats, duplicates pile up, stale notes drown the good ones. Mnueron is the opposite. A nightly reflective pass reads your store, merges duplicates, retires superseded statuses, and writes a log of what changed. Recall stays sharp as your data grows.
- 1Reflective pass. Scheduled low-cost LLM read across your store finds duplicate clusters, dated relative time references, and entries superseded by newer ones.
- 2Review queue. Proposals land in your dashboard. Approve, reject, or flip auto-apply on once you trust it.
- 3Audit log. Every merge, delete, and rewrite is captured in a tagged memory you can recall by date — nothing happens behind your back.
- 4Local or hosted. Free local users run
mnueron consolidatemanually. Paid users get it scheduled across every machine, with sync running on the same cycle.
mnueron · consolidation-log3:14 AM
Nightly run · 2026-05-25
namespaces touched .. 4
entries deleted ..... 22
exact-dup pairs ..... 2
time-fixes .......... 0
saved as ............ mnueron:7ba69936
patterns surfaced:
- elevizio ns ~80% redundant
chunks across 6 sessions
- default ns full of dev
test entries (14 deleted)
- scheduled-task prompts
becoming memoriesNext passtomorrow · 3:00 AM
How we compare
mnueron vs. the rest of the memory stack
Most memory engines were built for cloud-only agent runtimes. mnueron meets developers where they already work — Claude Code, Cursor, Cline, Windsurf — and keeps your bytes on your machine by default.
| Feature | mnueron | Mem0 | Letta | Zep | OpenAI Memory |
|---|---|---|---|---|---|
| Local-first / offline mode | Yes | Partial | Partial | No | No |
| Open source | MIT | Apache | Apache | Apache (core) | No |
| MCP-native IDE integration | 10 IDEs | No | No | No | No |
| Session continuity (auto-resume) | Yes | No | Partial | No | No |
| Web chat capture (Claude/ChatGPT) | Yes | No | No | No | ChatGPT only |
| Hybrid search (FTS + vector) | Yes | Yes | Yes | Yes | Yes |
| Auto-synopsis on save | Yes | No | Partial | No | Partial |
| Fact extraction | Yes | Yes | Partial | Yes | No |
| Entity resolution (cross-session) | Yes | Yes | Yes | Yes | No |
| Knowledge graph | Yes | Yes | Partial | Yes | No |
| Procedural memory (runbooks) | Yes | No | No | No | No |
| Self-revising memory (LLM merge) | Yes | No | Yes | No | Partial |
| Temporal reasoning (bi-temporal) | Yes | Partial | Partial | Yes | No |
| Webhooks (HMAC-signed) | Yes | No | No | Partial | No |
| Secret redaction on write | Yes | No | No | No | No |
| BYOK (your own AI key) | Yes | No | No | No | No |
| Free unlimited tier | Local | Partial | Self-host | Partial | Capped |
Where mnueron leads today
- Local-first by default — your memories never leave your machine unless you opt in.
- MCP-native — auto-wires into Claude Code, Cursor, Cline, Windsurf, Continue, Zed, Aider, Goose, OpenCode, and Codex.
- Web chat capture — Chrome extension scrapes Claude, ChatGPT, and Gemini conversations.
- Bring-your-own AI key — pay nothing extra if you already have a Claude or OpenAI subscription.
- Server-side secret redaction — 13 regex patterns scrub keys, tokens, and PII before persisting.
- Provider-agnostic — works with any LLM. No lock-in to a single model vendor.
Engine roadmap (honest)
Where competitors lead today and where we're heading.
- Knowledge graph — Auto-extract entities and relationships from memories — like Mem0's graph mode and Zep's Graphiti. Our fact extraction is the first stepping stone.
- Temporal reasoning — Time-aware relationships ("changed jobs in March") on top of our date-range filters.
- Cross-session entity resolution — Recognize that "John" in memory #4 is the same "John" in memory #200, without explicit tagging.
Comparison reflects publicly documented capabilities of each project as of May 2026. We may have missed nuance — open an issue on GitHub and we'll update.
Pricing
Free forever locally. Pay only when you want sync.
Local mode is unlimited and free. Pick the paid tier that matches how much you sync.
Free Local
Full local mode: MCP server, dashboard, Chrome extension, SDKs. No account.
$0forever
- All six memory tools
- Hybrid keyword + semantic search
- Chrome extension + history backfill
- Plugin system + redaction
- Self-host the backend yourself
Basic
Persistent memory for individual developers. Cross-machine sync + AI extraction.
$10/month
- Everything in Free Local
- 50,000 memories synced
- All major AI tool integrations
- Chrome extension + history backfill
- Email support
Popular
Pro
Advanced memory and productivity tools for power users and creators.
$19.99/month
- Everything in Basic
- 150,000 memories (3× Basic)
- Advanced search + ranking
- Automation tools
- 12,500 LLM calls/month
- Priority support
Team
Shared workspaces, role-based access, audit log. Built for teams that build together.
$49/month
- Everything in Pro
- Shared team workspaces
- Role-based access control
- Audit log + retention policies
- Unlimited team members
Enterprise (SSO, BAA, on-prem) — contact us.
Your AI's memory belongs on your machine.
Install in one command. No credit card. Open source.