Know what every work item actually cost.
Your agent can tell you what one session cost. It can't tell you what a ticket cost across the five sessions it actually took. traceknot won't let that slide: it stops each session before it runs, makes you name the work, then rolls every session against it into one real total — entirely on your machine.
Spent this week
342 sessions
Tokens
1.6M in · 0.8M out
Coverage
share of sessions claimed
Unattributed
3 sessions to triage
Top work items by cost
Top models by cost
Top agents by cost
Cost over time
Recent sessions
Sample data, for illustration
Works with
Every session starts with a question
traceknot could silently guess which ticket you're on. It doesn't — and it doesn't just log the spend quietly in the background either. It's a guardrail: your agent's first prompt is intercepted until you name the work, on purpose, so the record is something you actually stand behind, not a best guess stitched together after the fact.
Attach Context
Match this session to a work item, or skip
Target prompt
Fix the token refresh race condition in the auth retry path
Sample data, for illustration
It asks, it doesn't guess
No AI classifier quietly assigning your work. You pick the item, every session.
It actually pauses
A hook early in your agent's session holds it there until you claim a work item or skip — 240 seconds to decide.
Becomes a habit
Naming the work before the agent runs turns into muscle memory, not a chore.
This runs as a hook wired into your agent CLI at install time —see exactly how it's set up.
Local, and open about it
traceknot doesn't ask you to trust a vendor with your session data. It runs on your machine, and the code that runs it is public.
Everything stays local
Your session data lives in a database on your own machine, with a dashboard only your machine can reach. No account, no cloud sync, no analytics.
Read-only, by design
Issue-tracker lookups reuse the GitHub, GitLab, or Jira CLI you already have signed in — they read an issue's key and title, and never create, edit, or comment on anything.
Open source
Apache-2.0 licensed. Every hook, normalizer, and pricing table is on GitHub for anyone to read or audit.
Go deeper
Each of these gets its own page.
Install
One command sets up a local daemon, per-agent hooks, and an optional analysis skill.
Read moreHow tracking works
What gets recorded for every session, and what conclusions you can actually draw from it.
Read moreSession analysis
Export a session or work item and ask your own agent why it happened.
Read more