ax - the agent experience layer<br>a feedback loop for your coding agentTurn every agent session<br>into a better next run.<br>Built because we got tired of guessing what actually works.<br>ax watches every session your coding harness runs, spots the mistakes it repeats, and turns them into small, repo-specific fixes you review and apply — one at a time.<br>install in 30 secondsCopy setup promptRead the docs<br>paste it - your agent installs ax, labels your skills, and tells you which ones to actually use✓ Copied - paste into your coding agent for the guided setup
the loop, in your terminalYour sessions, working for the next one.<br>Every session lands in one local graph your agent can reach back into — to catch a repeat mistake, replay a clean run, recover the moment a test fails, or grade the harness itself. Same loop, whatever you run it in.
retroreplayrecalldoctorclassify<br>● live<br>Claude CodeOpus 4.8 · ~/Projects/ax
open the studioMission Control for your agents.<br>Run ax serve and the whole HUD lights up — archetype sigil, live activity, streaks, token spend by model. The wrapped highlights pop out as ax finds them.
127.0.0.1:1738
local timelive · 312 active days<br>23:1407<br>Tuesday<br>JUN 16, 2026 · 2,847 sessions traced
archetype · high<br>The Cartographer
archetype · primaryhigh confidence
The Cartographer<br>Maps the whole repo before touching a line — reads wide, edits once.
sessions<br>2,847<br>58.2K messages
tokens<br>1.4B<br>+18% vs prior 90d
activity · dailylive
312 active days21 weeks
streak<br>23d
best 41 days
peak hour<br>11 PM<br>most active
model split · 365d~$4.2K total<br>opus46% · $1.9K
sonnet31% · $1.3K
fable14% · $588
haiku6% · $252
gpt-53% · $126
wrapped
$ when you shipTuesday nights
wrapped
$ paths covered73%
wrapped
$ cost vs rewardlands cheap
Run ax serve to see yours — Mission Control, the Improve deck, Agent Wrapped, sessions and skill triage.
Used by engineers from
Segment
Segment
the pipeline · what feeds the graph<br>auto · paused<br>claude transcripts<br>~/.claude/projects
codex sessions<br>~/.codex/sessions
git + hook fires<br>repo + ~/.claude/hooks
typed local graph<br>surrealdb · 127.0.0.1<br>4,773 sessions · 369,132 turns<br>FTS median 5.9ms
interventionsranked · safety-contracted · brief-only128 applied → agents
every event → typed graph → ranked interventions. local SurrealDB at 127.0.0.1. see /features for the schema →six forms: skill · guidance · subagent · hook · automation · harness_check
open sourceIf it shapes your agent, you should be able to fork it.<br>AGPL-3.0. Local-first. Typed end to end. Every signal lives in files and a database you own — no cloud, no lock-in. Commercial license available if you need it.
~/code<br>$gh repo clone Necmttn/ax<br>Cloned ax into ./ax<br>$cd ax && bun install<br>1 284 packages installed in 4.2s<br>$ax daemon start<br>ax dashboard → http://127.0.0.1:1738
AGPL-3.0<br>Free & open · commercial license available
TypeScript<br>End-to-end, strictly typed
Local-first<br>SurrealDB on 127.0.0.1
Forkable<br>Hack the loop, ship a PR
Star on GitHub →Fork the repo →Join the Discord →Read the docs →
for teamsFree for your loop. Evidence for your team.<br>Everything above runs on your laptop, yours to fork. When you need to see how a whole team ships with AI — what’s spreading, what’s stuck, and what should become standard practice — that’s ax for teams.
See ax for teams→
01Install ax →02Routing →03Features →04Docs →
For teams·Registry·Brand<br>AGPL-3.0 licensed · 2026 · the missing layer