The hosted home for agent-trace

Make your code rich in context

For every AI-written line, the open-source agent-trace CLI records which model and coding tool wrote it — and the prompt and session transcript behind it. The intent that never survives into the code itself, kept next to every line, readable by both your team and your coding agents.

See attribution in the browser

Browse files with per-line blame, open the conversation behind any line, and see how much of your repo each model wrote — all from TracesHub.

Per-line blame with model, tool, and the prompt behind every AI-written line.
Repo-wide analytics — each model's share of every line at the default-branch tip.

Install the CLI

curl -fsSL https://traceshub.com/install.sh | bash

Python 3.9+ and zero external dependencies. See the CLI documentation for every command.

How it works

1

Install and init

agent-trace init

Zero-prompt setup in any git repo, just like git init. Hooks are wired up locally; your working tree stays untouched.

2

Code with your agent

git commit

Work normally with Cursor, Claude Code, or Codex CLI. On commit, the post-commit hook links traces to the commit and builds the per-line attribution ledger.

3

Push to TracesHub

agent-trace push

Add your TracesHub remote once, then push explicitly — git-style, with control over the level of detail that ships. You get a browsable web view of line-level AI authorship, shareable when you want it.

Context your coding agents can read back

Code alone can't tell an agent why it exists. With agent-trace, the model, tool, prompt, and full conversation behind every line are one command away — for you and for the next agent that touches the file.

agent-trace blame src/app.py

Per-line answers from the ledger: AI-written or not, and which model and coding tool wrote it.

agent-trace context src/app.py -l 10-50

Fetch the prompt and session transcript behind any line range — summaries and previews by default, full transcripts on demand.

agent-trace rule add context-for-agents -t claude

Install a prebuilt rule into Cursor, Claude Code, or Codex so your agent looks up this context on its own before editing.

Full command reference in the CLI documentation.

For you — and the whole team

The CLI records the context; TracesHub makes it browsable and shareable. Use it solo on your personal account as your own hosted record, or connect your GitHub org and give everyone — reviewers, new teammates, auditors — the full story of the codebase.

Install the GitHub App

Install on your personal account or your GitHub organization, then import any repo as a TracesHub project in a couple of clicks.

Bring the whole team

Organizations with member roles, per-project collaborators, and invitations — share a single project or the entire org.

One shared view of the project

Everyone browses the same record: per-line blame, the conversation behind each line, and repo-wide analytics of what each model wrote — no local setup needed to look.

Push safely from anywhere

Each developer mints their own scoped token to push traces from their machine — revocable any time, scoped to the projects it should touch.

Provenance for AI-written code

Deterministic, not guesswork

Attribution is built at commit time from per-line SHA-256 hashes. A line is AI-written or it has no attribution — there is no heuristic scoring, ever.

Local-first, shared by choice

Hooks write traces to ~/.agent-trace on your machine — nothing lands in your working tree, and nothing syncs until an explicit push. Config toggles and push filters decide what ships: full conversations, prompts and summaries, or just the attribution.

Git-like flow

agent-trace init, push, pull, sync. Attribution also travels as git notes, so it follows the repo through fetch and push.

Works with your agents

Traces AI-generated changes across Cursor, Claude Code, and Codex CLI — one attribution ledger no matter which agent wrote the code.

Blame you can trust

agent-trace blame tells you, per line, whether it was AI-written and by which model and tool — with a browser file viewer that overlays git blame and AI authorship.

The intent behind the code

Every trace links lines back to the prompt and full session transcript that produced them — the “why” that is never inferable from the code alone.

Open source at the core

TracesHub is built on standalone open-source components you can run and self-host yourself — the hosted product is the collaboration layer on top.

Start tracing in minutes

Install the CLI, run agent-trace init, and push your first trace — browse it yourself, and invite your team when you're ready.

Get started for free