Every capability, one context engine.
↓ scroll
Sources
Seven kinds of artifacts, one ingest pipeline. Each lands in a project space as embedded chunks with provenance.
One graph, every source
Code + docs + designs + meetings + tickets + PR threads — searchable as one.
Learn moreHeading-aware chunker
Token-budgeted chunks sized for precision retrieval without losing meaning.
Learn moreIncremental indexing
Push a commit, get fresh chunks. No full re-index for one-line changes.
Learn moreRoles & audiences
PMs, devs, QAs, designers, and clients are first-class users — same product, different windows onto the same graph.
Agents & automation
Read spec → write code → run tests → security scan → open PR. End-to-end agent pipeline that you can audit and replay.
Drift detection
When a spec doc moves after the code shipped, ContextBrain flags it — same day, not next quarter.
Architecture graph
Visualize modules, dependencies, blast radius — pre-rendered server-side for instant load.
Surfaces
Same context pack, every surface. Build a new surface as a thin layer over the same primitive.
Grounded chat
SSE streaming, role-aware system prompts, audience-filtered retrievers, inline citations.
Learn moreEmbed widget
Drop-in <script> tag for any PM app. HMAC user-identity or per-project share-token.
Learn moreClaude Code MCP
Stdio + SSE MCP server. Your editor speaks the same protocol as the rest of the platform.
Learn moreMemory & learning
ContextBrain learns from every agent run, every chat correction, every code change. The corpus grows itself — humans approve, never type from scratch.
Automatic fact extraction
Decisions, patterns, and constraints get extracted from chat + agent runs, queued for one-click approval.
Learn moreThree-tier memory
Session → project → org promotion happens automatically when an extracted fact recurs across enough distinct sessions.
Learn moreHybrid retrieval
Semantic + keyword + entity fused with RRF, then cross-encoder reranked. Same retriever powers chat, agents, and MCP.
Learn moreEntity-linked memory
Each memory resolves to the symbols + tickets it talks about. The graph shows what a refactor will invalidate.
Learn moreTime-aware memories
Freshness + supersedes lineage means a stale memory loses to a fresh one, and you can trace the chain back.
Learn moreCross-encoder reranker
Slots in after RRF and before freshness. Local BGE or hosted Cohere — pluggable via env.
Learn morePII governance
Every memory write is scanned for secrets, PII, and external email. Org admins can hard-block on match.
Learn morePublic eval numbers
Recall@10, MRR, nDCG, factual accuracy, tier correctness. Run on every release. Updated at /memory.
Learn moreTrust & security
Procurement-ready posture: encryption, RBAC, audit, BYOK, self-host. Per-org configurable.
Audit-grade trust
Every action logged, exportable to CSV. PII auto-redaction on Anthropic / OpenAI / AWS / GitHub keys.
Learn moreBring your own keys
BYO OpenAI / Voyage embeddings, BYO Anthropic agents. No data flows through our accounts.
Learn moreSelf-host or SaaS
Apache-2.0 core. docker compose up. Same code on your laptop and in production.
Learn moreSee the matrix in motion.
Connect a repo and walk the surfaces — no credit card.