XTrace
Sign inGet started
Stanford
Draper

BACKED BY STANFORD & DRAPER

Your team runs hundreds of AI sessions.
None learn from each other.

Your team runs hundreds of AI sessions.
None learn from each other.

XTrace turns what worked into steps people and agents reuse.

Start free

Book a working session

−82%

lower cost on work your team has already done once

63%

of sessions start from what the team already worked out

3 weeks → 3 hours

for a new hire to get up to speed

Day 1

new people work the way your best person works

See, learn, do. It’s that simple.

Your team’s work flows in. What worked gets kept and mistakes learned. What’s learned becomes something any person or AI can use.

01 Observe02 Distill03 Share04 Automate
Observe where the work already happens.A brain for the work your team does.Share what your AI learned with the rest of your team.With the right context and know-how, AI becomes an extension of your team.

Sessions

Everything your people and agents did today, captured where it happened.

Today

Arun DesaiRewrote the ingest retry policy after the Friday incidentClaude Code14m ago
Mia ParkPositioning pass on the Q3 enterprise pitchClaude1h ago
Sam OkaforNorthwind renewal call — pricing objection, security reviewGranola3h ago
Arun Desaimemory-hub · PR #512 — brains folder paginationGitHub5h ago
CaptureClaude Code plugin · Chrome extension · MCP connectors · direct upload — nobody changes how they work.

Repo: acme/payments-api

One brain per repo, team or account — holding what was decided, not what was said.

FACTARTIFACTEPISODE1,204 memories

Retries on 5xx are limited to idempotent methods; 429 always retries and honours Retry-After.

fact · from Claude Code session · 14m ago

Spec: Retry policyv4

artifact · versioned, diffable · governs app/retry.py

Friday incident → policy rewrite → PR #512 review

episode · 3 sessions, 2 people, 1 agent

Shared with

Platform team ·editGTM ·readSupport ·read

Sources

GitHubNotionSlackClaude Code
AskWhy did we cap the retry backoff at five seconds?Send

Answers cite the session, the call or the document they came from — so a claim can always be traced back to where it was made.

Share brain

Named teammates, at the access level you pick — not a link, not the whole workspace.

Repo: acme/payments-api2 selected
Priya RamanPriya Ramanpriya@acme.com✓
Sam OkaforSam Okaforsam@acme.com✓
Mia ParkMia Parkmia@acme.com

Access level

Viewer

Can read the brain and everything it fires.

Contributor

Can add sessions, specs and docs.

Admin

Full control — access, policy, delete.

Shared “acme/payments-api” with 2 teammates as can viewCancelShare brain

Who has access

Arun DesaiArun DesaiOwneradmin
Mia ParkMia Parkcan edit
Priya RamanPriya Ramancan view
Sam OkaforSam Okaforcan view

Access is per person, per brain. Anyone can leave their own grant.

Sales can read what engineering decided — without gaining access to the repo it came from.

Routines

The parts of the loop that repeat, running on a schedule or an event.

Monday GTM digestat 09:00, only on Monday · UTC

Summarise what engineering shipped last week from the repo brains, and write the customer-facing version into #gtm.

Succeeded · 2 days ago · 1 artifact

Review PRs against the specon pull request · acme/payments-api

Check every PR against the versioned spec artifact and comment where the implementation has drifted.

QueuedRunningSucceeded
Renewal risk alertevery day at 07:30 · UTC

Watch the account brains for objections that repeat, and alert the owner in Slack before the QBR.

DOC

Week of 27 Jul — what shipped, in customer language.docx

artifact · produced by Monday GTM digest · cites 9 memories

See how your team really uses AI.

Every session leaves a trail. You can see what people ask for, what they reuse, and where the same work gets redone. Set the guidance once and everyone’s agents pull toward the same goals.

Where the effort goes

Which teams, tools and brains get used, and which questions keep coming back week after week.

Guidance that sticks

Write the rule once as a skill. Every agent follows it, so the team stays on track without you checking each answer.

Spend you can explain

Token cost by team and by task, with repeat work you already paid for called out so you can turn it into a skill.

Team view

LAST 30 DAYS

1,284

AI sessions

63%

started from memory

−$8.4k

token spend

MOST REUSED THIS MONTH

Pricing renewal procedure

31 runs

Ship a migration safely

24 runs

Draft a security review

17 runs

Operations is redoing the same refund answer 40 times a week. Turn it into a skill.

Built for teams that run on AI.

Engineering

Six engineers, six agents, six different answers to the same question.

What goes wrong

Every Claude Code session starts cold. The spec someone wrote six weeks ago sits in a doc nobody reopens, so the code drifts away from it quietly and you only catch it in review, or in production.

What changes

One shared repo brain holds the specs, the decisions, the reviews and every past session. New sessions start from what the team already decided, and a routine checks each pull request against the spec that covers those files.

$/memhub:spec check

Spec: Retry policyv4· covers app/retry.py, app/**/backoff.py

! app/retry.py changed 3× since v4 · wait cap is now 30s, spec says 5s

→ suggest a change, or update the spec

Sales

The objection you just lost to was answered on a call three weeks ago.

What goes wrong

Everyone does their own research and keeps their own call notes. What the customer asked for, what you promised, and the worry that keeps coming up are spread across seven places and shared in none of them.

What changes

Calls, emails and chats turn into one page per customer that the whole team can search. When the same worry comes up again, the owner hears about it before the renewal call, not after it.

Northwind · account brain
318 memories · 4 people · updated 2h ago
FACT
Security review blocks the renewal until we attach SOC 2 proof. Raised on three separate calls.
Granola · 12 Jun, 2 Jul, 24 Jul
ARTIFACT
Objection handler: pricing vs. the incumbentv3
AlertRenewal risk routine → #sales-northwind, every day at 07:30 UTC
Operations

The playbook is in one person’s head, and they are on leave.

What goes wrong

How the work gets done lives with the person who has done it before: what comes first, the exception you always hit, who to call. Write it down once and it is out of date by next quarter.

What changes

The steps become a saved playbook your team and your AI can both follow. When someone does it a better way, the playbook gets a new version with the reason attached, so nobody is following an old copy.

How we run a customer escalation
skill · v6 · org-wide
SHARED
01Check how big the problem is before paging anyone.
02Open the incident channel. The CSM talks to the customer, not the engineer.
03Post the timeline to the account brain before the post-mortem, not after.
v6 · "CSM talks to the customer" added after the 18 Jul escalation · reason attached
Go-to-market

You are describing a product that shipped two quarters ago.

What goes wrong

What you say about the product drifts from what it does, because the people writing it are not in the room when it changes. Finding out means interrupting someone, and the answer comes back in language you cannot use.

What changes

The team that builds it shares what it knows into the same workspace, so a routine can read what shipped last week and write the customer version, with every claim traced back to where it came from.

Asked in Studio
"What changed in the product last quarter that Northwind asked for, and who promised it?"
Three of their five asks shipped. Rate limiting landed inpayments-api v2.4and Dana promised it on the 12 Jun call. SSO scoping slipped to Q4.
2 facts1 artifactGranola · 12 JunPR #488
DOC
Week of 27 Jul · what shipped, in customer language
made by a routine · cites 9 memories
FOR DEVELOPERS

The same memory, as an API.

Send the conversation and get back facts, files and session summaries, saved in your own index and scoped to a person, an agent or a shared group. No database to run.

SDK
npm install @xtraceai/memory
MCP
claude mcp add xtrace
ingestsearchrecall
import { MemoryClient } from '@xtraceai/memory';

const client = new MemoryClient({
  apiKey: process.env.XTRACE_API_KEY,
});

const job = await client.memories.ingest({
  messages: turns,
  user_id: 'alice',
  conv_id: 'conv_2026_08_01',
});

Questions that come up first