HN Explorer

Show HN: Traces: A new way to share and discover agent traces (traces.com)

13 points by nwpwr ↗ HN
Traces is a new way to share and discover agent traces. You can setup a personal or team account, and share publicly or privately (in a team).

We use Traces internally to capture and share our agent conversations. Every PR has a Traces link attached to it. We even have skills that can automate that for you (run 'traces setup' after install).

You might ask:

1) Why would I share traces? Well, we have found ourselves wanting to learn from each other on how to prompt different models and agents. We built Traces as a tool for teams to learn that together, and for us to learn that as an open community.

2) What about Privacy? You can share publicly, privately, or via direct link. We strip data at source to make sure anything that looks sensitive never leaves your machines. Orgs have more privacy settings as well. But honestly, me saying this is one thing, earning your continued trust is the goal of every release.

If you're interested, just get started at traces.com. You can hop into our Discord, find us on X at x.com/tarunsachdeva / x.com/tracesdotcom.

Thank you!

7 comments

[ 3.1 ms ] story [ 23.6 ms ] thread
hankyone
I like this idea of attaching a trace to a PR. Just makes sense now if you're going to have an AI agent write your code
millingab
Just tried it, its pretty cool. I love the UX overall. The product seems free so far. How are you planning on supporting it over time?

Are there plans for monetization by selling the agent trace data? In that case, I would recommend providing a paid subscription for people who don't want their data used for training.

sarimmalik
I've been really enjoying sharing traces to share prompting insights with the rest of our team. It helps the product is also beautifully designed.
andrewnez
I'm been trying this out the past couple weeks and found it very handy to share a trace with my team to give more context along side a pull request.

Can't wait for it to integrate more automatically with my git workflow too.

MeTa4
This is pretty amazing. In multi-agent systems traces become part of the working state of the system itself. If earlier agents produce a structured trace of their reasoning, tool calls, and intermediate results, downstream agents don’t need to rediscover the problem from scratch. A shared reasoning ledger between agents.
danialhasan
Love this! Once you adopt the primitive you can't go back
lakshyaag
Amazing use case and much needed! I wonder if down the line, it would be possible to allow for handoffs between different agent products (Cursor -> Claude Code, Codex -> Pi, for example)