Is this a vector store or RAG system?

No. Facts are typed edges (subject, predicate, object) with a status and a confidence score, not embeddings ranked by similarity. A predicate like decided_to_use means something specific; it is not a fuzzy match.

What happens if two tools capture conflicting facts?

They surface as a conflict you resolve, not a silent overwrite. The older fact stays on record as historical rather than disappearing.

Who can see what I capture?

Every fact has an exposure level: visible on your public identity, shared with a project's team, or private to you. You set it per fact, not per account.

Is Arachnest open source or can I self-host it?

No, Arachnest is closed-source, hosted software. It's currently invite-only while it's in alpha.

What happens to my work if I'm removed from a project?

You lose access to see and change it from then on. What you already contributed stays on record; it is not erased.

Which AI tools does it actually work with?

Claude Code over MCP for reading and writing as you work, a browser extension for ChatGPT, Gemini and Claude, and direct GitHub repository import.

Alpha

A memory graph for the decisions behind your work.

Arachnest keeps one structured record of what you and your projects have decided, why, and on whose say-so. Your AI tools read from it over MCP instead of each keeping a separate, forgettable history. Not a vector store. A typed graph where every fact points back to the conversation it came from.

Create account Sign in

Built and run by a small number of people on their own machines so far. Expect rough edges.

billing-svcdecided_to_usePostgres

status: confirmed
decided by: you, in session 3
evidence: turns 9 to 13
exposure: project

Typed, not embedded
Facts are subject-predicate-object edges with a real status, not similarity scores in a vector index.
Traceable
Every fact points back to the exact conversation and turns it came from. Open it months later and read the moment.
Invite-only for now
Arachnest is in closed alpha, access is by invite while we get the foundations right.

Capture, structure, then trace back.

1
Capture from where the work happens
A coding session in Claude Code, a thread in ChatGPT or Gemini, a pasted transcript, a GitHub repository. Arachnest pulls candidate facts out of the real exchange rather than asking you to write notes by hand.
2
Store it as a typed graph
Each fact becomes an edge: a subject, a predicate, an object. Beside it sits a status of confirmed, tentative, rejected or historical, a confidence score, and a record of who decided it. Contradicting decisions surface as conflicts, not silent overwrites.
3
Follow any fact to its evidence
Every edge carries a pointer to the exact conversation and the range of turns that support it. Open a decision and read the moment it was made, months later.

Capture happens wherever you work.

Every surface writes into the same graph under your account, so context does not fragment by tool.

MCP
Coding agents such as Claude Code connect over the Model Context Protocol and both read context and record decisions as they work.
Browser extension
Capture a ChatGPT, Gemini or Claude thread, or inject relevant context back into a fresh one, from a Manifest V3 extension.
Paste and import
Drop in a raw transcript, or point Arachnest at a GitHub repository to seed a project from its README and history.

ChatGPT

The popup on a supported chat page. One click sends the open conversation to your graph, or pulls relevant context back into it.

What a fact looks like in the graph?

Entities are nodes. Facts are typed edges between them. Access is per project.

nodes: People, projects, and the concepts they touch. The same concept named by two people resolves to one node. Separately tracked projects never merge on their own; an explicit invite the other person accepts is what links them.
edges: Typed. A predicate like decided_to_use means something specific, not a similarity score. Ambiguous captures wait in a review queue instead of being guessed at.
access: Per project. Being removed takes away what you can see and change from then on, without erasing the work you already contributed.

youperson

billing-svcproject

Postgresconcept

ownsdepends_ondecided_to_use

status: confirmedevidence: session 3, turns 9-13

One edge, fully specified: typed, scored, and traceable to the moment it was decided.

Questions worth answering directly.

Is this a vector store or RAG system?: No. Facts are typed edges (subject, predicate, object) with a status and a confidence score, not embeddings ranked by similarity. A predicate like decided_to_use means something specific; it is not a fuzzy match.
What happens if two tools capture conflicting facts?: They surface as a conflict you resolve, not a silent overwrite. The older fact stays on record as historical rather than disappearing.
Who can see what I capture?: Every fact has an exposure level: visible on your public identity, shared with a project's team, or private to you. You set it per fact, not per account.
Is Arachnest open source or can I self-host it?: No, Arachnest is closed-source, hosted software. It's currently invite-only while it's in alpha.
What happens to my work if I'm removed from a project?: You lose access to see and change it from then on. What you already contributed stays on record; it is not erased.
Which AI tools does it actually work with?: Claude Code over MCP for reading and writing as you work, a browser extension for ChatGPT, Gemini and Claude, and direct GitHub repository import.

Start with one project.

Create an account, connect one tool, and let the first few decisions land. It is alpha software: run it for something real, not yet something fragile.