compare
recall vs a CLAUDE.md file
A CLAUDE.md file is the right place to start. It stops being enough when the team, the file and the number of agents grow. This page says where, and what recall adds.
what it does well
What a rules file does well.
For short, stable rules, a file in the repo is hard to beat.
- It is plain text in the repo, reviewed and versioned like code.
- The agent reads it at the start of every session, with no service in between.
- It costs nothing and needs no account.
- Build commands, style, where things live: rules that rarely change fit it well.
where it breaks
Where it breaks.
"The CLAUDE.md is 600 lines and nobody reads it, including the agent."
It grows.
Every correction becomes another line. The whole file is read every session, so each line costs context whether it applies to the task or not.
It goes stale.
A line has no date and no author on its face. Nothing tells the agent that a rule written in March was reversed in May.
No reasons or conditions.
A rule without its reason gets applied where it does not fit, or dropped when it looks odd. There is no place for "holds when".
No outcomes.
Nothing records whether a rule held, was reversed or was regretted, so a bad rule reads exactly like a good one.
No per-agent limits.
Every agent that reads the file reads all of it, and any agent that can edit the repo can rewrite the rules.
Context, not a check.
The file sits in context. Nothing compares the agent's plan with it before the tool call runs.
what recall adds
What recall adds.
The same rules, kept as records a person can date, question and settle.
| Question | A CLAUDE.md file | recall |
|---|---|---|
| Where did this rule come from? | A line in a file; git blame, if anyone looks | Who said it, a person or the agent, with the quote and the date |
| Why, and when does it hold? | Unstructured text, if someone wrote it | because and holds when on every record |
| Two rules disagree | Both lines stay; the agent picks one | A person settles it; a model never erases either |
| Did the rule work? | Not recorded | Held, reversed or regretted; reversed ones are flagged in every later brief |
| What may each agent read and write? | All of it, plus edit rights to the file | Grants with a ceiling; no agent can write a rule that binds a person |
| Before the agent acts | The file is in context | Plan check, before the tool call |
| What the agent reads each session | The whole file | A brief, then what matches each prompt |
| Cost and setup | No cost, in the repo | A service; early access by invitation |
use both
Keep the file. Move the decisions.
Build commands and conventions that never change stay in the CLAUDE.md. Decisions, their reasons and their exceptions go to recall, where a person can settle them.
01Does recall replace a CLAUDE.md file?
No. Keep it for short, stable rules. recall holds the decisions that change, with the reason, the condition and the outcome.
02Can recall read the CLAUDE.md I already have?
A space can be taught from a Markdown document. What it learns from a document is attributed to that document, not to a person.
03Does the agent still read everything every session?
No. It gets a brief when the session starts, then what matches each prompt, and plan check before it acts.
04Is there a number that shows the difference?
Not yet. The repeat-rate test, a CLAUDE.md file vs claude-mem vs recall on the same corrected mistakes, is not measured yet. It will be published on /trust with its denominator.
early access
Connect, once you have access.
Early access is by invitation while we load-test capacity and cost. With access, every MCP client uses one endpoint.
https://app.recall.hiretrainva.com/mcp