Your team's agents should stop rediscovering work.
ACE gives Claude Code, Codex, and MCP agents a private team library of solved setup paths, traps, fixes, and receipts — so the next session starts from what your team already learned.
One team, one isolated ACE.
Your team's lessons stay in your team's ACE instance. Public sharing is optional, deliberate, and only for portable lessons.
Works where agents work.
ACE is MCP-native. The same retrieval idea supports Claude Code, Codex, and other MCP clients as teams adopt them.
Advice with evidence.
Capsules carry the trap, the first move, and verification context. Agents treat them as advice, not commands.
Every agent session forgets what the team learned.
Team ACE is for small teams already using coding agents heavily.
One developer's Claude session figures out a framework trap. Another teammate's Codex session hits the same class of problem a week later. Without a shared memory layer, the second agent starts cold.
ACE turns those lessons into short reusable capsules and makes them searchable at task start.
- Setup paths: the exact boring steps that worked after the docs led nowhere.
- Dead ends: approaches your agents should not waste another run on.
- Fix receipts: sanitized evidence, versions, commands, and verification notes.
- Team-local retrieval: private lessons stay private while generic gotchas can use Public ACE.
How Team ACE works without enterprise bloat.
Your agent checks team memory
At task start, the agent searches your private ACE instance for relevant team-shared capsules.
→A teammate submits a lesson
Useful session context becomes a cleaned draft: no raw transcript dump, no executable payloads.
→An admin approves it
Team-shared capsules go through review before other agents can retrieve them.
Private team loop first. Public network second.
Your team's work stays in your team's instance.
Team ACE is the proof engine: private setup, private lessons, private retrieval. It is aimed at SMB teams, agencies, and AI-heavy dev shops — not a full enterprise suite yet.
private team pilotsThe public exchange still matters.
Public ACE remains the global library for generic agent/devtool gotchas. Teams can use it alongside their private library, and later choose what is safe to contribute back.
public library liveWho should try this now
Team ACE is in private pilot mode.
Best fit: a small engineering team already running Claude Code, Codex, or MCP-capable agents on real work, and willing to test whether shared agent memory changes a real task outcome.
Not the pitch: quantified token savings, SSO, on-prem, or enterprise compliance. Those come later if the private team loop proves useful.