yoru · open source

Know what your agent did, and whether it was safe.

yoru records what your coding agent actually does: every tool call, file edit, and red-flag. Then it gives the session a read across throughput, reliability, and safety, so you open it already knowing if it needs review. A short Claude Code hook, post-event, with a 2-second timeout, so a yoru outage never stalls your terminal. Public beta; self-host, or run the free single-developer cloud.

Get started

See it on GitHub

Open source · CLI MIT, server AGPL · Claude Code hook, local-first · public beta

Works with

Claude Code today · Cursor and Aider write the same transcript shape (alignment in progress)

What it does.

Every tool call, recorded

A Claude Code hook captures each tool call, file edit, and prompt as the session runs, so there is a timeline of what the agent actually did, not a guess after the fact.

Red flags at ingest

yoru flags risky patterns as events land: secrets, rm, destructive SQL, .env writes, migrations, CI changes. The dangerous moves surface instead of hiding in a wall of activity.

A read on the session

Each session gets a grade across throughput, reliability, and safety: a single legible read, so you can tell which sessions need a closer look instead of paging through every trace.

Local-first, self-host

The CLI is MIT, the server is AGPL. Run it on your own box and your session data stays on your network. The cloud is free for one developer.

Install the hook. Your next session shows up.

Install the CLI and run init; the Claude Code hook is wired and your next session appears in the dashboard. Post-event and timeout-guarded, so it never blocks your terminal. yoru is in public beta and actively being reworked, so expect rough edges.

pip install yoru-cli   # then: yoru init

GitHub ↗yoru.sh ↗

Straight answers.

Does it slow my agent down?: No. The hook is post-event: it fires after Claude Code finishes a step, not during, and the POST runs with a 2-second timeout, so a yoru outage never stalls your terminal.
How is this different from Langfuse or Helicone?: Those trace the LLM calls a product makes to serve its customers. yoru records the coding-agent sessions your developers run to build the product: different unit, different buyer. You would use both.
Where does my data go?: To the yoru instance you run. Self-host keeps it on your network; the CLI only POSTs to the server you set. The code is public, grep the repo.

Follow the build

When we ship or learn something real about watching coding agents in production. No filler, no fixed schedule.

Running agents unattended across a team?

yoru is how we keep an eye on our own agent sessions. If your team is running coding agents at scale and needs to see what they did and whether it was safe, that's the work we do.

Talk to us

Know what your agent did, and whether it was safe.

Open source · CLI MIT, server AGPL · Claude Code hook, local-first · public beta

What it does.

Every tool call, recorded

A Claude Code hook captures each tool call, file edit, and prompt as the session runs, so there is a timeline of what the agent actually did, not a guess after the fact.

Red flags at ingest

yoru flags risky patterns as events land: secrets, rm, destructive SQL, .env writes, migrations, CI changes. The dangerous moves surface instead of hiding in a wall of activity.

A read on the session

Each session gets a grade across throughput, reliability, and safety: a single legible read, so you can tell which sessions need a closer look instead of paging through every trace.

Local-first, self-host

The CLI is MIT, the server is AGPL. Run it on your own box and your session data stays on your network. The cloud is free for one developer.

Install the hook. Your next session shows up.

pip install yoru-cli # then: yoru init

Straight answers.

Does it slow my agent down?

No. The hook is post-event: it fires after Claude Code finishes a step, not during, and the POST runs with a 2-second timeout, so a yoru outage never stalls your terminal.

How is this different from Langfuse or Helicone?

Those trace the LLM calls a product makes to serve its customers. yoru records the coding-agent sessions your developers run to build the product: different unit, different buyer. You would use both.

Where does my data go?

To the yoru instance you run. Self-host keeps it on your network; the CLI only POSTs to the server you set. The code is public, grep the repo.

Follow the build

When we ship or learn something real about watching coding agents in production. No filler, no fixed schedule.