Learn it. Keep it. Share it. What one agent learns, every agent and teammate knows.
Two ways in
Plug it in or build on it.
Memory Cloud + MCP
Use it
For anyone on Claude or Codex. Install the MCP and share context with coworkers and their agents in one command. No code required.
Install once, works in Claude and Codex
Share context with a coworker or their agent in one line
Bring any team memory straight into a chat
# add xtrace to Claude or Codex
$ claude mcp add xtrace
Procedural Memory API
Build on it
For teams building their own agents. Give any agent memory of the tool calls, paths, and fixes that worked, Implement in a few lines.
Capture & complete procedures deterministically
save() and load() agent memory from the cloud
Framework-agnostic: vercel-ai-sdk, pydantic-ai, or custom
# add memory to your own agent
$ npm install xtrace
BUILT ON FULL TRAJECTORIES
Memory companies store facts. XTrace captures the whole trajectory — every tool call, path, and outcome — so your agents learn procedures that actually worked, and you can see the receipts.
PROCEDURE IN YOUR MEMORY CLOUD
LEARNED
ship-figma-feature.runbook
Learned from 3 accepted runs · shared by Sarah · used by 4 teammates
1
Read the Figma file, extract components
2
Match against the design system in components/
3
Open a PR, run the review loop
4
Merge only after CI and one review round pass
Share with a teammate
Edit steps
Delete
SOURCE RUN · JUL 14
ACCEPTED
00:04
✓
figma.get_file
xtrace123abc
00:11
✓
grep
components/ --design-tokens
00:37
✓
github.open_pr
#142
01:52
✓
ci.review_loop
2 rounds
02:58
✓
github.merge
#142
✓ Accepted by Sarah — this run became the procedure
+ 2 earlier accepted runs fed this procedure
FIRST TIME, WITHOUT THE PROCEDURE
11 turns · 14.2k tokens
EVERY RUN AFTER, WITH IT
2 turns · 2.5k tokens
−82%
tokens per task
Share skills, artifacts, SPECs and memory.
procedural memory
Build a landing page with TypeScript, host on Vercel, add dark mode, and use isometric graphics…
Saved those steps as a reusable skill. Share it with anyone in one click.
Upload to Memhub
Share with Alice
artifacts
Draft the company brain spec
Company brain spec
Doc · Markdown
Share with Bob
Share with Codex
Whether your product is a split chat-and-canvas workspace or a sidebar agent hub, it is still an AI-native app that needs memory and comms.
Agentic App
Agents
Integrations
Dashboard
Your Agent Name
What would you like to do?
XTrace API / SDK
AI-Native Software
Something here
Do this and do that
Roger this and that!
Tool calls
Ask AI…
Vercel AI SDK
Pydantic AI
Hermes Agents
And more!
Procedural memory
XTrace embeds and indexes what your agent knows, so every task starts with instant similarity search — not another pass through a folder of markdown.
VS
Stop shipping endless agent skills
No bespoke skill for every workflow — XTrace gives your agent shared comms and memory it reuses across tasks.
No more DIY file systems
Stop bolting on storage so clients can dump markdown everywhere. Structured memory lives in XTrace, not scattered docs.
Keep the context window clean
Retrieve only what matters at request time instead of stuffing an agent's entire history into every prompt.