fluidrecall.com · for any AI agent
Practice in Muse with a custom connector
Muse can write a custom connector to Fluid Recall today from the public skill and OpenAPI document. We have not submitted a directory connector. Practice stays in the Muse chat.
Directory connectors and custom connectors
Muse uses the word “connector” for two different things. A directory connector is submitted at Muse’s platform, reviewed by Meta, and listed under Settings so any person can connect it with a tap. A custom connector is one Muse writes for a single person when the service is not in that list. Meta says it does not review custom connectors. Credentials go into Muse’s Secure Credentials Store.
Fluid Recall ships for the custom-connector path. We have not submitted a directory connector. That work is parked on purpose. The public skill and HTTP API are still written so a later directory review could test them end to end, and so any other agent can call the same operations today.
What to point Muse at
Ask Muse to connect Fluid Recall as a custom connector. Give it the public documents, not a localhost URL. Muse’s VM cannot see a process on the person’s laptop.
- The imperative contract: https://fluidrecall.com/muse/skill.md
- The HTTP schema: https://fluidrecall.com/api/openapi.json
- The Markdown doors: https://fluidrecall.com/api/openapi/surfaces.json
New integrations prefer POST /api/agent/v1/session/start with no Authorization header. That mints a personal token and returns the first due prompt as Markdown. Later calls send Authorization: Bearer <token>. They do not send a user id. The token is the tenant. A copy in the chat is fine. Keeping it in Secure Credentials is also fine.
Reads, important reads, and writes
Muse’s safety model treats retrieval and mutation as different classes. Fluid Recall splits the tools the same way so a custom connector — and a later directory reviewer — can see the classes.
Routine reads: list decks, list what is due, list the cards in a deck, read one card’s prompt. Important read: reveal one answer after the person has tried to recall. Writes: create a deck, add a card, correct a card, retire a card, record a rating. recordReview is the only write that moves dueAt. The due list never includes answers.
Practice stays in the chat
This website does not run study drills. After Muse has a token it should speak one prompt, wait for the person’s attempt, reveal that one card, send a rating, and trust the dueAt in the response. Details are on the spaced practice page and in the skill. If you are a person, copy the starter prompt and paste it into Muse. Do not look for a signup form here.
The same contract for any agent
Muse has distribution, and other agents are likely to mirror its connector shape. That is why the skill is public and why the API is plain HTTP instead of a localhost plugin or an MCP-only door. A Claude, ChatGPT, or terminal agent that can POST JSON to a public URL can run the same loop. Fluid Recall does not require MCP.
Questions
Is Fluid Recall in the Muse directory?
No. Directory submission is parked. A person asks Muse to build a custom connector and points it at the public skill and OpenAPI document.
Does practice leave the Muse chat?
No. Muse is the coach. Fluid Recall is the system of record. This website has no study screen and no signup form.
What if I do not use Muse?
Use the same skill and HTTP API. Prefer POST /api/agent/v1/session/start, then the Markdown surfaces. The personal token is the tenant. Do not send a user id.