Compare
Mossbear does two things: it delivers the rules you already write — AGENTS.md, CLAUDE.md,
.cursorrules, and four more formats — into every repo and agent that reads them, and it
grades whether a session actually followed them, against criteria you wrote yourself. Most
tools that look adjacent do one of those and not the other.
What makes mossbear different?
Section titled “What makes mossbear different?”| Category | Examples | What they do | What mossbear adds |
|---|---|---|---|
| Rule-sync tools | rulesync, ruler | Keep one rules file in sync across repos and tools | The same sync, plus a verdict on whether a session followed the rule |
| Eval and observability platforms | Promptfoo, Braintrust, Langfuse | Score model outputs or trace agent activity, usually in CI or a separate dashboard | Grades sessions against guides you wrote, as you work, not a rubric shipped by us |
| Agent memory and retrieval layers | Mem0, Zep | Extract facts into a vector or graph store and retrieve them automatically, on every turn | Writes plain files into the repo you already track in git by default; further retrieval, where it exists, is a query your agent asks for, not automatic ranking |
| Team skills-governance platforms | Enterprise skill/rule governance tools | Score and gate agent skills for a team, managed by an admin | Grades against guides you write yourself; runs without an admin console |
| Notes and knowledge tools | Obsidian, Notion | Where you write things down | Reads the rule files you already have; doesn’t ask you to write anywhere new |
Grading needs hooks, and hooks are wired per agent. Guide delivery works everywhere in the table above; exactly which agents also get action logging and grading today is on the local agents page.
Rule-sync tools
Section titled “Rule-sync tools”If you already keep a rules file in sync with a symlink or a small script, that part is
solved and mossbear does not argue with it. mossbear link reads the same files —
AGENTS.md, CLAUDE.md, .cursorrules, .agents/, and the rest — and keeps them in sync the
same way. What a sync tool cannot tell you is whether the rule did anything: a symlinked
file that every agent ignores looks identical to one they all follow. Mossbear grades each
session against the guide, so you get a verdict instead of a copy.
Eval and observability platforms
Section titled “Eval and observability platforms”These tools score model outputs against a rubric, usually in CI, or trace and log what an agent did, usually in a dashboard aimed at a team debugging production traces. Both are real jobs mossbear doesn’t try to replace. The difference is what gets graded and against what: mossbear grades a whole agent session in your own repo against the guides you wrote for it, continuously, as part of your normal workflow rather than a separate step.
Agent memory and retrieval layers
Section titled “Agent memory and retrieval layers”These store extracted facts in a vector or graph database and retrieve them into an agent’s
context automatically, on every turn, based on inferred relevance. Mossbear’s default path
is different: guides materialize as plain Markdown files in a repo you already track in
git, the same way AGENTS.md or CLAUDE.md already work — nothing to operate yourself, no
embedding pipeline, no index to rebuild. Mossbear does hold a context store for notes and
documents, reachable through an MCP pull_bundle tool, but
retrieval there is a query your agent asks for, not automatic relevance ranking run on
every turn.
Team skills-governance platforms
Section titled “Team skills-governance platforms”Some products score and gate the skills or rules a team installs, enforced by an admin across everyone’s repos. That is a different buyer and a different job: mossbear grades your own agent against guides you wrote, runs locally, and needs no admin console or central policy to work for one developer on one machine.
Notes and knowledge tools
Section titled “Notes and knowledge tools”Obsidian, Notion, and similar tools are where people already keep their notes, and mossbear doesn’t try to replace that either — it reads the rule files that live in your repo, the ones your agent already reads at the start of a session, rather than asking you to write instructions somewhere new.
Trying it
Section titled “Trying it”The CLI is open source (MIT) and free to install. A dashboard account to see graded sessions is also free, no card required — see getting started.