Every AI agent starts from zero.
Yours doesn't have to.
> Give every agent and every person on your team one shared memory, in plain files your company owns.
OPEN SOURCE · 4K+ STARS · 25K+ DOWNLOADS/MO · FREE TO START
Own your memory · 764 notes · 2,420 relations
Models are rented.
Your knowledge is the asset.
The model behind your agents will change every few months, and so will the tools around it. What matters is what your team knows: what you decided, why, and what you've already tried. Today that knowledge lives in fragmented chat histories and vendor memory.
It belongs in one stable, always-accessible place your company owns, that every agent and every person works from.
Basic Memory is where you keep what you know, and where agents and people leave context for each other.
“Basic Memory provides the one thing AI models on their own are still missing: consistent, persistent working memory.”
Not sure it fits? Paste this into the AI you already use.
Read https://basicmemory.com/llms.txt. Then ask me a few questions about how I use AI in my work. Based on my answers, tell me whether Basic Memory would help, which parts I'd actually use, and where it wouldn't be a good fit.
Works in Claude, ChatGPT, or any agent that can read a web page.
No more re-explaining
New agents, new sessions, and new hires start with the context your team has already built.
Work you can check
Agents work from notes people can open, so anyone can see where an answer came from.
Knowledge that survives change
Swap models, tools, or vendors. Move people between projects. The memory stays.
Owned, with a predictable bill
Plain Markdown you can export anytime, on a flat plan with a spending limit.
Same question. A different AI.
The answer is still there.
A launch decision gets saved in a Claude chat. Days later, a fresh Codex session is asked about it.
Why did we move the Harbor launch to March 12?
I don't have any context about the Harbor launch or a date change. Could you share the project details, or the discussion where this was decided?
You moved it from February 26 to March 12 so the billing migration could finish first. That was decided in the January 14 planning meeting. The launch checklist still says February 26. Want me to update it?
read from decisions/harbor-launch-date.md
---title: Harbor launch datetype: decision---# Harbor launch date## Observations- [decision] Launch moves to March 12 #harbor- [reason] Billing migration must finish first- [source] January 14 planning meeting## Relations- part_of [[Harbor Launch]]- blocks [[Launch Checklist]]Example exchange. The answer comes from a Markdown file you can open, edit, and keep.
“With this tool I have solid, revisitable records of the ideas and decisions made, whenever I need them. And I am not locked into one AI service, I just connect Basic Memory to everything I use!”
Memory that belongs to you,
not to your tools.
Your AI tools will keep changing. The knowledge you build with them should outlast every one of them, and stay something you and your team can read.
“What makes Basic Memory compelling to us is that memory stays inspectable outside the AI itself: portable, machine-readable, and understandable by a human. That makes it feel less like hidden model state and more like durable shared infrastructure between a person and an agent.”
Your knowledge, as a workspace.
Kept current by agents.
With 1.0, your knowledge becomes a workspace your team can read, with a wiki and live notes, and keep current, with agents whose work you can see and review.
Wiki
Open a project as a wiki: an overview that links into its decisions, research, and people, so anyone on the team can find their way around.
Browse a project the way people read
Live notes
Put a task board, chart, KPI, or timeline inside any note. It's written in Markdown and pulls from your other notes.
Task boards, charts, timelines, and more than 30 other views
Documents
Bring in PDFs, Word files, slides, and CSVs. Each one becomes a searchable note, and the original file stays one click away so you can check the source.
PDF · Word · PowerPoint · CSV
Agents Beta
Give an agent instructions in a note. It runs when you ask or when new notes land in a folder, and every change is yours to inspect, keep, or undo.
Included on paid plans · billed by usage credits
Plain files, with real search on top.
A folder of Markdown is easy to own and hard to search. Basic Memory search helps your AI find all the relevant information, not just the notes with matching words.
Semantic search
Find by meaning, not just keywords. Vector and hybrid search across everything you've written.
Structured search
Filter by frontmatter. Status, type, tags, priority, or any custom field. Query your notes like records.
Schemas
Define and validate note types. Task, Person, Client, Ticket. So structure holds as the graph grows.
Knowledge graph
Typed relations and observations connect notes into a graph your AI can actually traverse.
Less reading, faster answers.
In a folder of notes, like an Obsidian vault, an AI reads whole files to find one fact. Basic Memory lets it search first and read only what it needs, which saves time and tokens.
Same question: “What did we decide about Q4 pricing, and why?”
A folder of notes
- 1Greps for “pricing” and gets dozens of matches
- 2Opens several whole files to check which one matters
- 3Misses the note titled “Plans for next year”: different words
- 4Next session, starts over
Basic Memory
- 1One search, by meaning: the decision and the meeting it came from
- 2Reads the section it needs, not the whole file
- 3Follows a link to who decided and what changed since
- 4Next session, the same search finds the same answer
Already keep notes in Obsidian? Basic Memory works on your vault →
“I chose this solution because it is not over-engineered: it provides what I need, it does not assume what I need or force me into some weird workflow. There is no blackbox - my memory stays observable, stored in local Markdown files that I can backup or edit on my own.”
Works with the tools you already use: Claude · ChatGPT · Codex · Cursor · Obsidian · VS Code · any MCP client
Who makes Basic Memory?
Basic Memory is made by Basic Machines: founder Paul Hernandez with cofounders Drew Cain and Drew Smith.
We built Basic Memory because we think it needs to exist. As AI takes on more of the work, people should stay in charge of it: working with AI, not having AI work for them. That starts with a memory you can read, shape, and own.
We put support first: we're easy to reach and quick to respond, and we keep making Basic Memory better for you and your team. We use it every day, and we plan to keep building it for a long time.
“Basic Memory changed my whole relationship with LLMs... I am completely revamping all our companies processes around a Basic Memory workflow.”
Start your knowledge graph.
Free and open source to run on your own machine. Cloud plans from $19/month, with a 7-day trial. Your files, your graph, forever.
Read the docs →Partner program for MSPs & agencies →
Keep building your knowledge.
Product updates, practical guides, and ideas for working with Basic Memory.