Finds what you meant
Search by the words you remember or by what you meant. It finds the note either way, offline.
Memory for you
and your agents.
Your agents forget everything when a session ends. Memorinth remembers: what you decided, what you learned, and why.
Explore
This is a cortex.
37 notes about Memorinth, written the way you and your agents write them. Click one to open it, or ask it something.
Interact with the cortex
Group nodes by
Inside a cortex
A folder of Markdown, with a shape.
A cortex is plain files on your disk. Your notes stay as you write them. Next to them, three folders hold what your agents learn, each with one rule, so the memory stays trustworthy as it grows.
Your notes
Projects, specs, journals: everything you write, in any folder you like.
Agents read them for context and never change a word.
Projects/Payments/Checkout flow.md
memory/
What happened and what was learned: a bug's real cause, a preference, a how-to. Dated, in the agent's own words.
So the next session starts where the last one ended.
The login bug was a time zone mix-up
decisions/
Every choice and the reason for it. A decision is never edited: a new one replaces it, and the old one stays, marked superseded.
So an agent doesn't undo a choice it never heard of, and you can see how you got here.
0004 Answers from a local model or Claude · replaces 0003
facts/
Stable truths about your work, one per note, each with its source. Every fact comes up for a recheck after 180 days.
So agents act on what's true now, not on what was true last spring.
Runs on Windows, macOS and Linux · source: release.yml
Features
Built for thinking with agents.
Answers with sources
Ask your cortex a question. Every claim links to the note it came from.
Every link a synapse
See your cortex as a living graph, and watch an answer light up the notes it used.
Terminals built in
Run Claude Code beside your notes and watch the cortex change as it works.
Dictation, offline
Speak a note. It's written down on your machine, and nothing is sent anywhere.
Memory that stays true
Facts carry a source and come up for a recheck. Decisions are replaced, never rewritten, so you see what changed and why.
Your notes stay yours
Agents write their own folders. Everything else in your cortex they only read.
MCP
Your agents remember, wherever you work.
While an agent works on your project, it talks to Memorinth over MCP. Before it starts, it looks up what it should know. When it's done, it saves what it learned. Every tool that speaks MCP shares the same cortex, on your machine.
MemorinthMCP server · your cortex
- Claude
- Codex
- Gemini CLI
- Cursor
- and any MCP client
Your project~/code/shop
Looks up before it starts
Each session begins with a briefing of current facts and decisions, and the agent searches the cortex before it touches what it doesn't know.
Saves as it works
It keeps what it learned, records decisions with their reasons and adds facts with a source. You don't have to ask.
One memory, every tool
Move from the terminal to your editor and the memory comes with you. Every note says which agent wrote it.
Connect in one step
Click Connect Claude Code in Settings, or run one command. Other tools add memorinth-mcp to their MCP settings.
$ memorinth-mcp setup claude-code --cortex ~/Memorinth ✓ MCP server added ✓ Session briefing hook installed ✓ Save reminder hook installed
The app
Your notes, your agents, one window.
Local first
Your cortex is a folder of Markdown files. Search, the graph and dictation run on your machine.
Plain files
Open them in any editor, put them in git, take them anywhere. Memorinth adds no lock-in.
You choose what leaves
Notes go to a model only when you ask Claude a question. Use Ollama and nothing leaves at all.