DocuMemory turns your AI conversations into clear, searchable documentation that keeps what you decided, why, and what worked. Claude, ChatGPT, Cursor and other tools all save to it and read from it, so nothing gets lost or stuck in one app.
Private & Encrypted 15 free distillations, no card MCP Native
MEMORY ENGINE PIPELINE READY
MEMORYADRcapture → distil → recall
WORKS WITH THE AI YOU ALREADY USE
ClaudeClaude CodeChatGPTCursorWindsurf+ Any MCP Client
01 / WHY DOCUMEMORY
You worked it out with AI. Then it scrolled away.
AI is part of most people's workday now, by choice or not. It helps you reach real answers, but those answers end up in chat threads that were never meant to be a record.
Docs nobody has time to write
Everyone wants documentation when they pick up work someone else touched last. Almost nobody has the time to write it.
Answers buried in old chats
Threads kept open just in case, and days of scrolling back to find the one command that finally worked.
Nothing carries over
What you settled in ChatGPT is invisible to Claude, and every new session starts from zero.
So hand the writing to your AI, and keep it where every AI can read it.
Your AI writes the documentation. Say “save this to DocuMemory” and the conversation becomes a clear document: what you decided, why, and the final working result.
Every AI tool shares one memory. Claude, ChatGPT, Cursor and others all save to and search the same place, so your knowledge isn't locked to one vendor.
You can write it yourself, too. Add your own guides, notes, and rules for your AI tools to search and follow.
WHY I BUILT IT
“I use several AI tools at work, and the useful parts of those conversations kept getting lost. I'd leave old threads open just in case, dig through days of history for one command, and start from scratch every time I switched tools.
What I wanted was for my agents to write the documentation for me, and for all of them to read it back from one place that doesn't belong to any single vendor. That's DocuMemory.”
The maker of DocuMemory
BUILT FORDevelopersIT & operationsData & analyticsProduct teamsSupport teamsAnyone who works things out with AI
02 / HOW IT WORKS
From conversation to lasting knowledge.
Keep the insight. Lose the noise. Give every new session a head start.
When a chat lands on something worth keeping, ask your AI to save it to DocuMemory. You never leave the conversation. You can also write or paste documents yourself in the dashboard.
save_chat_as_doc({ content, project_name })
03 / SIGNAL OVER NOISE
A long conversation. A lasting memory.
Your next AI session doesn't need every attempt that failed. It needs what worked, why it worked, and the rules worth remembering.
Documents
Answer “what did we decide, and why?” Decisions, guides, snippets, and notes, written for a person reading months later.
Memories
Answer “what's true right now?” Short facts and rules, plus a snippet any AI can load at the start of a session.
session.logEXAMPLE
# Search architecture discussion
You: Exact function names aren't showing up.
AI: Let's try increasing the vector weight.
You: That still misses exact identifiers.
AI: We could combine vector and full-text search.
You: The scores have different scales though.
AI: Use reciprocal rank fusion on each ranked list.
You: Yes. HNSW for pgvector, and scope to userId.
You: Save this to DocuMemory as a memory.
As it appears in your dashboardEXAMPLE
MemoryPlatform·
Search architecture and ranking rules
postgrespgvectorsearch
AI context snippet
Paste into another model's system prompt
### Project Memory: Search architecture and ranking rules
Scope: the platform's search layer (indexing, ranking, and query scoping).
- PostgreSQL + pgvector with an HNSW index for vector similarity.
- Semantic and full-text search run independently.
- Fuse the two rankings with RRF; never blend raw scores.
- Scope every query to the authenticated user.
Apply this by: treating each line as a constraint when changing search, ranking, or index code.
How search works across the platform: two independent search arms whose rankings are fused, never blended. Read this before changing ranking, indexing, or how queries are scoped.
Key facts
Search runs on PostgreSQL with pgvector, using an HNSW index for vector similarity.
Semantic (vector) search and full-text search run as independent arms.
The two arms are combined with reciprocal rank fusion (RRF), never a weighted score blend.
Every search query is scoped to the authenticated user.
postgres · pgvector · search
search-architecture-and-ranking-rules.mdEXAMPLE
---
title: "Search architecture and ranking rules"
type: MEMORY
project: "Platform"
tags: ["postgres", "pgvector", "search"]
summary: "How search works across the platform: two independent search arms whose rankings are fused, never blended. Read this before changing ranking, indexing, or how queries are scoped."
created: 2026-10-14T12:00:00.000Z
source: "mcp:memory"
generator: documemory
---
# Search architecture and ranking rules
> How search works across the platform: two independent search arms whose rankings are fused, never blended. Read this before changing ranking, indexing, or how queries are scoped.
## Key facts
- Search runs on PostgreSQL with pgvector, using an HNSW index for vector similarity.
- Semantic (vector) search and full-text search run as independent arms.
- The two arms are combined with reciprocal rank fusion (RRF), never a weighted score blend.
- Every search query is scoped to the authenticated user.
## AI context snippet
Paste this into another model's system prompt.
````markdown
### Project Memory: Search architecture and ranking rules
Scope: the platform's search layer (indexing, ranking, and query scoping).
- PostgreSQL + pgvector with an HNSW index for vector similarity.
- Semantic and full-text search run independently.
- Fuse the two rankings with RRF; never blend raw scores.
- Scope every query to the authenticated user.
Apply this by: treating each line as a constraint when changing search, ranking, or index code.
````
---
`postgres` · `pgvector` · `search`
04 / JUST ASK
No commands to learn.
Once DocuMemory is connected, talk to your AI the way you already do. It saves when you ask, and searches or loads what you've kept when that helps.
Everything it saves shows up in your dashboard, where you can read, edit, organize into projects, or delete it.
Save this discussion to DocuMemory.
A document with the problem, what you decided and why, and the final working result.
save_chat_as_doc
Save these Platform project conventions to DocuMemory as a memory.
A memory: the facts and rules that hold true, ready for any AI to load later.
save_memory
Check DocuMemory for how we handled rate limiting.
Your AI looks up what you already decided before it suggests something new.
search_docs
Load the Platform project's context from DocuMemory before we start.
Every memory for that project, brought into the chat at once.
load_project_context
Write a setup guide for our Docker config and save it to DocuMemory.
Saved exactly as your AI writes it, with no summarizing step.
create_doc
05 / WORKS WITH YOUR TOOLS
Connect the AI you already use.
Pick the one you work in. The easiest options take a minute and need no technical steps at all; developer tools are there when you want them.
Claude and ChatGPT, on the web and your phone
Add DocuMemory as a connector on claude.ai or chatgpt.com and sign in. No files and no tokens, and it carries over to their mobile apps. ChatGPT needs a plan with Developer mode.
EasiestNo token needed
Claude Desktop
Paste a short configuration into Claude's settings file and reopen the app.
A few minutes
ChatGPT custom GPT
Build your own GPT that saves to and searches DocuMemory through Actions.
A few minutes
Claude Code
One command in your terminal, and DocuMemory is there in every coding session.
For developers
Cursor and other MCP tools
Add a small entry to your MCP configuration. Other MCP-compatible tools, such as Windsurf, connect with the same address and token.
For developers
REST API
Save and search from scripts, automations, and CI jobs with a bearer token.
For developers
After you sign up, Settings → Integrations walks you through each option step by step, with your own details already filled in.
A second brain for everything you work out with AI.
01
Chats become documentation
Stay in Claude, ChatGPT, or Cursor. One request turns a conversation into a document that lasts.
chat-to-doc distillation
02
Keeps what worked
The final result and the reasoning behind it stay. The dead ends and detours don't.
zero-fluff code
03
One memory for every AI
Any connected tool can save, search, and recall, so switching tools never means starting over.
MCP native
04
Finds it by meaning or exact words
Search understands what you meant and still matches the exact name, command, or error you typed.
hybrid vector + full-text · RRF
05
Context ready for your next chat
Compact snippets give a new session the facts it needs without the transcript it doesn't.
prompt-injectable snippets
06
Write your own docs and rules
Add guides, notes, and rules for your AI tools to follow. Enhance with AI can tidy a rough draft.
Markdown · Enhance with AI
07
Share with your team
One shared space for a team, with each project open to everyone or limited to the people you choose.
organizations · per-project access
08
Private by default
Only you and the teammates you choose can see your workspace. Tokens are stored as one-way hashes, there are no ad trackers, and our AI providers don't train on your content.
hashed tokens · no ad trackers
09
Your data stays yours
Every document is plain Markdown you can edit, pull out through the API, or delete whenever you like.
Markdown · REST API
PRICING
Try it first. Subscribe when you're ready.
15 free distillations. No credit card required.
Free trial
US $0
15 uses, once per account
Explore your workspace and connect your AI client. Distillation and Enhance with AI share the free allowance.
Continue after your free uses with documents, AI integrations, and private uploads. Includes up to 200 distillations a month, counted from your billing date. Cancel renewal from Settings.
Continue after your free uses with documents, AI integrations, and private uploads. Includes up to 200 distillations a month, counted from your billing date. Cancel renewal from Settings.
per month, or US $150 per seat, billed annually · 2-seat minimum
One shared documentation repository for your team. Every member gets a seat; share projects with everyone or restrict them to specific people. Each member can run up to 500 distillations a month. No free trial: billing starts when you set up the organization.
DocuMemory is an MCP server over Streamable HTTP, with a REST API and an OpenAPI schema alongside. Clients authenticate with a per-client access token, or Claude and ChatGPT sign in over OAuth. Each connection acts in one workspace, with your permissions there.
No. If you use Claude or ChatGPT on the web or your phone, you connect by adding DocuMemory as a connector and signing in, with no code or configuration. Developer tools such as Claude Code, Cursor, and the REST API are there when you want them.
Does my AI save things on its own?
No. DocuMemory tells your AI to save only when you ask, such as “save this” or “remember that”. It may suggest saving when a conversation reaches a conclusion, but it waits for your yes. Your AI app's own settings can also require your approval before any tool runs. Everything saved appears in your dashboard, where you can edit or delete it.
Can I write or edit documents myself?
Yes. Write or paste documents in the dashboard, in a simple editor with headings, lists, checklists, and images (or in Markdown, if you prefer), and use Enhance with AI to tidy a rough draft into a clean format. Anything your AI saved can be edited the same way.
Is DocuMemory self-hosted or open source?
No. DocuMemory is a hosted service operated by Black Light Logic LLC, and its source code is not publicly available. You use it at documemory.dev and connect your AI tools to it; there's no server for you to run.
Which AI tools can connect?
Claude (Claude Code, the desktop app, the web, and the mobile apps), ChatGPT (as a connector on plans with Developer mode, or as a Custom GPT), Cursor, and other clients that support MCP over HTTP. Everything is also available through a REST API.
Which AI models process my content, and can I choose them?
Distillation runs on Anthropic's Claude, and semantic search uses OpenAI embeddings. The hosted service chooses the models, so you can't bring your own model or API keys. Both providers process content through their APIs, which don't use it for training by default. Details are in the Privacy Policy.
Who can see my documents?
Only you, in your personal workspace. In an organization, members see its shared projects, and owners and admins can restrict any project to specific people.
Can I get my data out, or delete it?
Yes. Every document is Markdown you can read and edit in the dashboard or pull out through the REST API. Deleting your account in Settings disables it immediately; it can be restored for 30 days and is then permanently deleted.
What counts as a distillation?
Each time DocuMemory turns a conversation into a document, or tidies a draft with Enhance with AI, that's one distillation. Searching, reading, and writing documents yourself don't count.
What does it cost?
Every account starts with 15 free distillations, no card required. After that, choose a monthly or annual plan, or per-seat billing for a team. Personal plans include up to 200 distillations a month, counted from your billing date. Team plans include up to 500 per member each month. See Pricing for current prices.