Bring bRRAIn into every AI surface your team already uses.
VS Code, Cursor, Claude Code CLI, Claude Desktop, the browser (Claude.ai / ChatGPT / Gemini / Perplexity), Copilot Chat — every shipping outbound client is auto-detected, configured, and rules-aware. Plus the MCP gateway, ingestion pipeline, integration policies, API tokens, and the embedded SDK.
Every AI surface, auto-configured.
Each client ships cryptographically signed and self-hostable. Auto-detects which AI assistants are present and writes per-AI configurations between bRRAIn-managed markers — your existing settings outside those markers stay untouched.
VS Code
Native VS Code extension. Bundles signed connectors for every supported platform — verified at install. Sidebar UI, built-in diagnostic command, OS-keychain credential storage.
- Auto-detects Copilot Chat, Cursor, Claude Code CLI
- Writes per-AI configurations + project rules files (CLAUDE.md, Cursor rules, Copilot instructions)
- Background fallback to capture turns the AI didn't volunteer
Cursor
Same extension package runs in Cursor with no separate install. Writes the appropriate Cursor configuration plus a bRRAIn-managed rules block — operator hand-edits outside the markers are preserved.
- One package, zero per-AI rebuild
- Operator hand-edits outside bRRAIn markers preserved
Claude Code CLI
Auto-detected when Claude Code is present. Wires the bRRAIn MCP server into the CLI configuration and pairs it with a CLAUDE.md workspace rules file, so context survives every invocation.
Claude Desktop
Native Claude Desktop extension, self-hosted from your own brain pod. Cryptographically signed tokens scoped per-client; integrity checksums published with every release.
Browser extension
Chrome, Edge, Safari. Four built-in adapters for the major web AI assistants: Claude.ai, ChatGPT, Gemini, Perplexity. Hybrid auto-capture with an always-visible manual button so a UI rev never breaks capture. Smart seeder + recent-captures panel + Ask-the-brain side panel.
/brrain start [project]— pre-fetch project context into the chat/brrain ask— bridge any web AI into the brain MCP gateway- Captures land in your vault in seconds, not hours
Copilot Chat fallback
When an AI client ignores its tool-call protocol, the VS Code extension's background hook captures the turns anyway — so context lands in the Vault even from non-cooperative assistants. Operator opt-out per client.
Standard protocol. bRRAIn-grade boundaries.
The MCP gateway is the canonical channel for AI tool integration. It exposes built-in servers, hosts a curated marketplace of vetted third-party servers, and routes everything through a bidirectional firewall with tiered inspection.
- HTTP bridge for browser-resident AIs that can't spawn local processes
- Forward-compatible wire format aligned with the open MCP standard
- Sensible safety limits on payload size and search scope
- Unknown server/tool combinations are rejected — no silent fallback
Pipe institutional knowledge in. Without losing the audit trail.
OAuth-connected providers, drag-and-drop file ingest, integration policies that constrain what gets through Zone 7. Every byte gets a vault path; every vault path gets a role tier; every role tier survives a backup-and-restore.
OAuth providers
Google Drive, Slack, Microsoft 365, Box, Dropbox, GitHub, Notion. OAuth credentials handled with strict in-memory and at-rest protections — never logged, never serialized to disk in cleartext.
File ingest queue
Drag-and-drop into the console with sensible safety limits. A persistent worker turns the queue over continuously, with live progress and animated transitions for every entry.
Integration policies
Per-provider redaction patterns, allowed file types, path globs, and action rules. Pre-loaded with sensible starter sets; same intuitive editor as security policies.
Zone 7 inspection
Every ingest payload goes through both security inspection gates. Content classification, PII detection, allowlist enforcement. Anything suspect ends up in Quarantine, not the Vault.
Webhooks
Cryptographically signed inbound webhooks for event-driven workflows. Idempotency keys ensure at-least-once delivery without duplicates — fire-and-forget for the producer, exactly-once for the Vault.
API tokens
Cryptographically signed tokens scoped per client. Hashed at rest; the plaintext secret is shown exactly once at mint. Last-used timestamps surface stale tokens for revocation.
Drop bRRAIn into your own apps.
The Go SDK is production-ready. Schema-agnostic ingestion, Service / Agent actor management, domain adapters for web frameworks (Gin, Echo, net/http, Express, Fastify, Flask, FastAPI), databases (PostgreSQL LISTEN/NOTIFY, MongoDB oplog), and message queues (Kafka, RabbitMQ).
Free Developer tier with approval. Volume-tier pricing for OEM partners and integrators. Third-party apps embed bRRAIn as their AI memory engine.
Read SDK docs →client := brrain.NewClient(apiKey) ws, _ := client.CreateWorkspace("acme-app") // Schema-agnostic ingest ws.Store(record) // Relationship-aware retrieval results, _ := ws.Retrieve(brrain.Query{ Search: "Q3 customer engagement", Mode: brrain.RelationshipGraph, })
Every outbound client is a trust boundary, not an integration.
Authentication, scope, sandboxing, and audit are part of the protocol — not an afterthought wrapped around it.
Cryptographic auth
Distinct, isolated authentication scopes for management, verification, and provisioning — a compromise of one does not authorize the others. Tokens hashed at rest.
Per-extension scoping
Each outbound client gets its own scoped token. Per-client revocation. Per-client rate limiting and policy enforcement on the roadmap.
Sandboxed execution
MCP servers run behind a bidirectional firewall. The Code Sandbox CVE-scans untrusted code; quarantine with feedback; nothing runs against the live Vault until it's been cleared.
Tamper-evident audit
Cryptographically chained per-zone audit log. Drop, reorder, or modify any event and the chain breaks detectably. The audit signing secret is persisted in the Vault, surviving process restarts.
Path-traversal blocked
Defense-in-depth path validation: every read goes through two independent rejection layers, with symlink escapes blocked at the filesystem boundary. Either layer alone is bypassable; both together are not.
Cross-platform support
Native binaries for every common operating system: macOS, Windows, and Linux on both Intel and ARM. Bundled into each client's release artifact and integrity-verified at install.
Pick your AI surface. We'll wire the rest.
Sign up and connect every shipping outbound client from inside your console — signed, integrity-verified, and ready in seconds.