alpha

One brain. Every assistant aligned.

Encrypted, role-aware knowledge for every AI assistant. taip serves your company's rules and context to every AI assistant over MCP, role-filtered before decryption.

Request a pilot

brain · context · tools

What that looks like in practice.

"One brain. Every assistant aligned." is what taip promises. This page shows what it means: what a rule looks like when it runs, what your assistant sees during a session, and what your auditor sees afterwards.

A rule in taip.

Here is an illustrative example of a single rule as taip serves it — not a real customer's data, but a realistic shape of what your own rules would look like.

Rule text > No patient identifiers in test fixtures.

Metadata

  • Severity: blocking
  • Pack: security

When an assistant is generating test data, taip serves this rule automatically as part of the session context. The assistant cites it mid-task — noting that it is avoiding patient identifiers in test data per the security pack. For context-mcp sessions, the access log records the read: actor, role, item, and timestamp.

Your team sees a consistent assistant that knows your company's requirements — without being reminded at the start of every session.

the product

What you actually get — in practice.

brain

Every rule, standard, and compliance pack your team writes once. Encrypted, versioned, and loaded by every assistant on demand — not restated in each new prompt.

context

Company knowledge that adapts to who is asking. Each assistant reads the slice its role permits — and nothing more — before any decryption happens.

tools

A curated MCP tool catalogue that publishes available actions to your AI assistant. taip publishes the catalogue — it does not enforce policy on tool calls or audit them. Your other tools work exactly as before.

getting started

Point. Define. Work.

Point your client at taip

Add your taip endpoint to your MCP config. Your existing tools — Cursor, Claude, GitHub Copilot, JetBrains — pick it up without changes to how they work.

Define who reads what

Set which roles can access which rules and context. Role filtering runs before decryption, checking plaintext frontmatter. Identity is established through tenant-level access control.

Work normally

Your assistants load the same encrypted rules every session. No re-explaining. No drift between tools or teammates.

trust & safety

The controls your auditor will ask about.

Encrypted at rest
Every rule body is encrypted before it leaves your tenant. Content is decrypted in memory only after a request is authorized — nothing readable remains at rest.
Role-checked before decryption
Authorization runs on unencrypted metadata before anything is decrypted. Role filtering checks plaintext frontmatter and fails closed. Identity is established through tenant-level access control — per-caller origin authentication is planned.
EU-hosted, one tenant per company
Each company runs in its own isolated environment, with its own encrypted store and keys — hosted in the EU, operated under EU law.

Read the full security overview →

feature status

What the product can do today.

CapabilityStatusNotes
Encryption at restLiveRule and context bodies are ciphertext. AES-256-GCM for packs, SOPS+age for the context repo. (C1)
Decryption in process memory onlyLiveNo plaintext written to disk. Hard constraint in taip-agent. (C2)
Role filtering before decryptionIn alphaFrontmatter RBAC runs per call and fails closed. Identity is asserted, not yet authenticated. (C3)
Authenticated origin auth and licence enforcementPlannedBuilt, but off in production. Only Cloudflare Access protects tenants today. (C4)
EU tenant isolationLiveOwn container, own encrypted repo, own keys, EU-hosted. (C5)
Audit log — context-mcpLiveEvery read and write logged with actor, role, item, timestamp. (C6)
Audit log — brainPlannedNot yet built. Beta blocker. (C7)
Same rules across every MCP clientLiveCursor, Copilot, JetBrains AI, Claude — any MCP client reads the same rules. (C13)
MCP tool catalogueLiveCurated catalogue of MCP tools, published to your assistants via the taip-tools proxy. (C14)
Policy enforcement on tool executionPlannedOut of scope. taip serves its own content — not a gateway over other tools. (C15)

Status as at 2026-07-29. Rows trace to claims-register.md §3.