⎿ 2 · Respect the 100/min rate limit set after the outage
⎿ 3 · Add per-user prefs — owned by Maya
✻grounded in 4 sources
Also available via MCP, CLI, API.
Batteries included
Foundation learns as you work
Foundation learns by following along with your team's work in Claude Code, Codex, Slack, and more. It finds both explicit and tacit knowledge of how work gets done and builds a self-improving wiki.
Real-time sync
Versioning & diffs
Lineage & citations
Concurrency control
Continuous background improvement
Incorporates human feedback
Foundation works everywhere
Foundation works in the tools your team already uses. We build and maintain all the connectors, data pipelines, and integrations for you. Bring your context with you to any model, harness or tool.
Sources
Claude Code
Codex
Cursor
Opencode
Slack
Notion
GitHub
Google Drive
Granola
Files
Destinations
Claude Code
Codex
Cursor
Opencode
Mac App
Web UI
Slackbot
CLI
MCP
API
Foundation is private and secure
Foundation is SOC 2 Type 2 and has granular access control policies to ensure private data stays private. For example, coding agent sessions stay private to the author.
Access control
SOC 2 Type 2
SSO / SAML
BYOC
Dedicated Support
SLA
Foundation is infrastructure, batteries-included, and endlessly-customizable
Foundation is built on ChromaDB and Context-1, our retrieval engine — the product of years of research and development. Foundation is SOTA on the BEAM memory benchmark, with first-class versioning, concurrency, lineage, and access control.
The real world is messy. Machine intelligence must be able to learn from experience and from the humans that teach it.
Foundation is our answer.
We believe that self-improving AI systems will change how all software is built and how organizations operate. Foundation is not yet perfect - but it's already very powerful.
Foundation is the culmination of many years of deep investment in technology at Chroma. It builds upon everything we've built to date: seminal research like Context Rot, a database purpose-built to store and search massive amounts of unstructured data, and custom post-trained language models to process and retrieve it.