LogRack is a shared pile of logs for you and your LLMs. A log is a typed piece of text: a task, an idea, a note, a piece of context. What Claude learns, ChatGPT can read. What you write in the portal, your coding agent picks up next session.
You run four AI tools. None of them know what the others did.
Every session starts empty. You paste the same background into a new chat, restate the same decisions, and hope you did not leave out the part that mattered.
What Claude remembers, ChatGPT cannot see. Your coding agent cannot see either. The context stays inside whichever tool you happened to be using that day.
So you keep your own notes as well. Now there are four versions of the plan, and nothing records which one is current or who wrote it.
One write path in. Two read paths out.
An agent writes a log over MCP. You write one in the portal. A script posts one over REST. Same pile either way.
The log is typed, tagged, filed to a rack, and stamped with the client that wrote it. Racks are created the first time something writes to one.
A context pack returns a paste-ready bundle inside a token budget. Catch-up returns only what changed since that client last looked. Nobody has to be told twice.
Six ideas. That is the whole product.
Everything you and your LLMs have written, searchable, in one place. Filing is optional. Search finds it either way.
Per-project piles. File logs by topic or leave them loose. A rack is created automatically the first time an agent writes to it.
Personal and team boundaries. A workspace holds its own pile, its own racks, and its own keys. Nothing crosses between them.
Every log records the client that wrote it. One API key per AI tool, and the key name becomes the writer. You can always tell Claude's note from ChatGPT's from your own.
Logs are never edited. A new version supersedes the old one and the old one stays. You get an honest history instead of a rewritten one. Nothing is deleted unless you delete it.
One call returns a token-budgeted markdown bundle you can paste straight into a prompt. Pinned logs first, then matches, then recent.
Written for agents. Readable by people.
One MCP server connects Claude, ChatGPT, and anything else that speaks MCP to the same pile. Tools for write, search, get, update, delete, context packs, catch-up, and racks.
An agent asks what happened since it last looked and gets exactly the delta. No re-reading the whole pile at the start of every session.
Full-text search across the pile. Free-form tags on any log. Logs typed as tasks carry open and done state, so one agent can file the work and another can close it.
Browse the pile, compose a log by hand, filter it down to what you need, and see stats on who wrote what.
For everything that is not an MCP client. The typed client is on npm as @lograck/api.
Postgres, a Next.js app, and one MCP binary. Run the whole thing on your own infrastructure.
One MCP server, one key per tool. The key name is how you tell who wrote what.
Google sign-in at app.lograck.ai. You get a workspace and a pile.
Create an API key for each AI tool and name it after the tool. That name becomes the writer on every log the tool writes.
Add the MCP server to each client with its own key. From then on they write to and read from the same pile.
yarn add @lograck/api Read the connect docs Sign in, connect your tools, and let them share what they already know.