Skip to content
MCP memory server for every AI

Every AI forgets.You shouldn't have to.

Your AI memory shouldn't be locked to one vendor. Capture a decision once, then recall it in Claude, ChatGPT, Grok, Gemini, Cursor, or any MCP client. You choose what is kept, and you can export or delete all of it anytime.

Free tier forever. Paste one URL into Claude. Cancel anytime.

See it work

Told Claude once.Now they all know.

Save a decision in one tool. Open any other, weeks later, and it is just there. One memory, no re-explaining, and you own every word of it.

You're in control

You choose what your AI remembers.

Not every project is your standard. Include the ones that are, and your AI only learns from your best work. Tap a project to move it in or out.

memory: 2 of 3 sources feeding your AI
Included2
Excluded1
Semantic recall

Ask in your own words.

OpenAI embeddings and Pinecone vectors with keyword fallback. Ask naturally, and memory answers with what you actually decided, scoped to you alone.

why did we switch to fetch + reset?
recalled · decision

We moved to fetch + reset so token revocation is one global lever (sessionVersion) instead of a per-session refresh to expire.

Tycoon Opsscoped to you
How it works

Capture, curate, recall.

01

Capture

Tell your AI to remember a decision or lesson while you work. One tool call, kept forever on paid plans, 30 days on Free.

02

Curate

You choose which projects feed recall. Nothing enters search unless you include it.

03

Recall

Weeks later, any AI asks memory first. Claude, ChatGPT, Grok, Gemini, Cursor, Claude Code: one brain.

Zero install to start: paste one URL into Claude as a custom connector on web, desktop, or phone. Not ready to sign up? Try the interactive demo first. CLI clients (Claude Code, Gemini CLI) connect with one command.

Personas

Your memory holds the standard your AI works to.

A persona is the bar your agent holds itself to: coverage, docs, error handling, review, speed. It lives in your memory and is recalled on every session, so the standard travels with you across Claude, Cursor, and Claude Code. Blend a few, dial the intensity, and the security floor never drops below 40%. Mix your own below.

The rigor ladder

2 in the blend

Effective standard

94%rigor and security
at 100% intensity

floor: 40%, never lower

Led by Enterprise Grade, your agent works to:

Test coverage
95 to 100%
Documentation
Full API docs, guides, changelogs
Error handling
Every edge case
Code review
Mandatory peer review
Priority
Quality over speed

For always-on agents

The agents that never sleep, finally remember.

An autonomous agent running on a server has no chat window to scroll back through. It restarts, it loses its context, and over its life it may be driven by more than one model. Lodekeep is the memory that survives all three: durable, curated, and shared across every model the agent runs.

Hermes, the always-on agent I run on a VPS, keeps its memory in Lodekeep. A restart, or a swap to a different model, never wipes what it knows.

Wire an agent up over the MCP bridge or the API with one scoped, revocable key. See how to connect.

hermes @ vps
$ systemctl restart hermes-agent  booting... memory in RAM: empty   → search_memory("what was I working on?")  ✓ recalled from Lodekeep    "deploy from main only"    "the staging DB is read-only"    "the client wants dark mode first"   context restored. resuming.
The obvious question

Why not just use what you already have?

Each AI already remembers things inside its own walls. mem0-style tools already capture everything. A markdown file already works and costs nothing. Here is the honest difference: vendor-neutral memory, deliberate curation, and data you actually own.

Vendor-neutral by design

One memory works across Claude, ChatGPT, Grok, Gemini, Cursor, and any MCP client. Switch assistants without leaving what you taught them behind.

Deliberate, not a firehose

You choose which projects feed recall, and every entry is yours to read. Nothing is captured behind your back.

Yours to keep

Export or delete everything, anytime, from your own dashboard. No lock-in, no black box.

Ready for teams

Start solo, then add a shared team memory pool when your whole team needs the same context.

Follows you across surfaces
Lodekeep
Yes. Same memory in Claude, ChatGPT, Grok, Gemini, Cursor, and any MCP client.
Claude's own memory
No. Stays inside the single app or project you set it in.
mem0 & auto-memory tools
Depends on the integration. Most wire into one app at a time.
A CLAUDE.md file
Only if you manually copy the file into every project.
You choose what gets remembered
Lodekeep
Yes. Per-project include and exclude toggle.
Claude's own memory
Limited. No per-project curation, just what the app decides to keep.
mem0 & auto-memory tools
No by default. Captures everything; you prune after the fact.
A CLAUDE.md file
Yes, because you typed every line yourself.
Semantic search over what you captured
Lodekeep
Yes.
Claude's own memory
Partial. Recent-context recall, not a searchable store.
mem0 & auto-memory tools
Yes.
A CLAUDE.md file
No. It is a flat file; Claude only reads what fits in context.
Own and export your data
Lodekeep
Yes. Export or delete anytime from your dashboard.
Claude's own memory
No. Lives inside the platform.
mem0 & auto-memory tools
Depends on the vendor.
A CLAUDE.md file
Yes. It was always your file.
Setup and upkeep
Lodekeep
Two environment variables. No manual upkeep.
Claude's own memory
Nothing to set up, but nothing to take with you either.
mem0 & auto-memory tools
You run or subscribe to another service.
A CLAUDE.md file
Nothing to set up, but it goes stale and never searches.

Fair reads, not strawmen: Claude's own memory is genuinely useful within its surface, auto-memory tools are a real convenience once you trust the firehose, and a markdown file costs nothing because you already own it. The gap is curation across every surface you actually use.

Security

Trust is the product.

Lodekeep holds your private context across every AI you use. That only works if you can trust it completely, so isolation, encryption, and ownership are first-class here, not fine print.

Isolated to you

Your memory lives in a namespace scoped to your account. No shared index, and one account can never read another.

Encrypted

TLS in transit, and every offsite backup is encrypted before it ever leaves the server.

Yours to keep

Export everything, or permanently delete it, any time. You own it, not the vendor.

Adversarially reviewed

Put through an intensive internal security audit before launch.

Pricing

Memory that pays for itself.

Start on Free, no card required. That is your trial. Upgrade to Pro, Max, or Team when your memory outgrows it. Cancel in one click.

If it saves you re-explaining your stack even a couple of times a week, it has already paid for itself.

Free

Free forever
$0

Try it on one real project.

  • 1 curated project
  • 25 searches / month
  • 50 captures / month
  • 30-day retention
  • Works across every AI
Start free

No card required. This is your trial.

Pro

For daily work
$12 /mo

Everyday memory for a working developer, across every AI you use.

  • 5 curated projects
  • 750 searches / month
  • 1,000 captures / month
  • Permanent retention
  • Persona blending
Get Pro

Max

Most power
$25 /mo

Unlimited memory, plus coding-agent clients and Obsidian vault sync.

  • Unlimited projects
  • Unlimited searches & captures
  • Permanent retention
  • Coding agents: Claude Code, Codex, Cursor, Gemini CLI
  • Obsidian vault sync
Get Max

Team

Shared memory
$49 /mo

Everything in Max, with one memory pool your whole team shares.

  • Everything in Max
  • Shared team memory pool
  • Add your teammates
  • Coding agents for the whole team
  • Obsidian vault sync
Get Team

A persona is a preset working style for your AI, from audit-ready Enterprise to move-fast Experimental. Blending mixes two or more, and every persona keeps the 40% rigor floor: the minimum level of testing and double-checking.

Payments by Stripe. Card details never touch our servers. Cancel anytime from the customer portal.

Questions

What everyone asks first.

Yes. Every row and every vector is scoped to your user id, and tool calls authenticate with your personal token. Nobody else can search your memory.

One brain, every AI

Give your AI its memory.

Stop re-explaining your codebase every session. Two minutes to install. Remembered forever on every paid plan.