The pixel-art mascot in a brass diving helmet swimming down through drifting transcript pages, holding a lantern.

hev kit

Take back your agency.

hev kit makes your coding agent traces searchable via a hybrid search system built on turbopuffer and hev layer. Every file your agents opened, every command they ran, every approach they abandoned sinks the moment the session closes. Install the hev daemon on as many machines as you want, and go back down for it.

$ hev up dive in ↓

Take back your agency.

hev kit makes your coding agent traces searchable via a hybrid search system built on turbopuffer and hev layer. Install the hev daemon on as many machines as you want to share and search your traces.

The hev kit dashboard: a search across coding-agent sessions, with matching prompts, replies and tool calls highlighted under each session.

twilight zone · 200 m

Ask it what happened.

Type what you remember. kit searches prompts, replies and tool calls across every session and shows where the words turned up, down to the command the agent ran.

Same search from the terminal: hev query "why did the preflight fail". Or just ask Claude Code or Codex: hev up gives them a skill that searches it for you.

Search results for 'why did the preflight fail', with matching prompts, replies and tool calls highlighted under each session.

Read it back the way it ran.

A trace replays in the harness's own shape: your prompt, the agent's narration, tool calls folded until you want them. The rail on the right marks every prompt, failure and wait.

This one is the session that wrote hev up.

A session transcript: the prompt, the agent's running narration with tool calls folded, and a timeline rail down the right edge.

See where the time went.

The timeline splits the wall clock into you, the model and its tools, with context size underneath. In that three-and-a-half hour session the agent spent 81% of it waiting on a person.

A gantt timeline with lanes for you, the model, tools and context, beside a table of where the wall-clock time went.

Know what it cost.

Spend, tokens and cache hits across every trace, by day and by model, at list rates. The heatmap shows when you actually prompt.

The stats view: total spend, tokens, traces and cache hit rate, with tokens and spend per day and a weekday-by-hour heatmap.

Run it on your own sessions.

$ brew install hev/tap/kit
$ export TURBOPUFFER_API_KEY=tpuf_...
$ hev up
  # gateway + dashboard in Docker, hevd under launchd,
  # and a hev-query skill for Claude Code and Codex
  # then open http://127.0.0.1:8099
  ✓ docker running
  ✓ layer-gateway:0.6.0 running on :8080
  ✓ turbopuffer key accepted, archiving to namespace hev-traces
  ✓ hevd installed, first scan started
  ✓ hev-query skill installed for claude, codex
  ✓ dashboard running http://127.0.0.1:8099
   ▄▀▄   ▄▀▄    hevd is up, capturing sessions on this machine.
  ▐████████▌    dashboard  http://127.0.0.1:8099
  ▐█ ▀  ▀ █▌    search     hev query "why did the preflight fail"
   ▀██▄▄██▀ ψ   archive    <(°O°)> turbopuffer · hev-traces
    ▐█  █▌      stop       hev down$ hev query "why did the preflight fail"

Needs macOS, Docker running, and a turbopuffer API key. hev down stops it all. Read the note on sensitive data before you run it.

What hev up starts

Gateway Docker
The hev layer community gateway, in front of your turbopuffer account. Hybrid search runs there, not in kit.
Dashboard Docker
Browse sessions, read a trace, search the archive, at http://127.0.0.1:8099.
hevd your Mac, launchd
The capture daemon. It reads the session files your coding agents already write and indexes new and growing ones, so the archive keeps filling after the terminal closes.hevd, a small red imp in a diving mask and flippers
hev-query skill ~/.claude, ~/.codex
An agent skill, so Claude Code and Codex can search your traces themselves: they plan a few phrasings, fan them out through hev query, and read the session around the best hits. Installed for whichever of the two you have.
Your archive turbopuffer
A namespace in your account, hev-traces by default. kit hosts nothing and keeps no copy.
hev/kit
Source, commands, configuration and the design RFCs Apache-2.0
Your transcripts may contain sensitive data.

Coding agent sessions hold whatever the agent saw: keys pasted into a prompt, secrets on a command line, file contents, customer data. hevd indexes every coding agent session on each machine it runs on, prompts, replies and tool calls, and writes them to your turbopuffer namespace as they are.

kit does not redact anything or filter by project yet. Run it on a machine whose sessions you are comfortable storing there, and treat the namespace and your key with the same care as the transcripts.