We built taip because we kept re-explaining our codebase to every new AI session.
The founder of Taip Systems OÜ operates medical software under the Medical Device Regulation and IEC 62304. By 2026 that codebase ran 138 production rules daily. Every new AI session — every new assistant, every new colleague — needed the same re-explanation. taip is that re-explanation, done once and served to every assistant.
brain · context · tools
What taip is.
taip has three surfaces, each addressing a different part of the re-explanation problem.
brain — Your rules, standards, and compliance context. Encrypted, versioned, and loaded by every AI assistant on demand over MCP. Write the rule once; every session starts with it.
context — Company knowledge that adapts to who is asking. Each assistant reads the slice its role permits, and that filtering runs before any decryption happens.
tools — A curated MCP tool catalogue your assistants can invoke. taip publishes the catalogue. It does not enforce policy on tool calls or act as a gateway over other tools.
The three surfaces are served to every MCP-capable client — Cursor, GitHub Copilot, JetBrains AI, Claude — without changes to how they work.
capability status
What runs today — and what we are building toward.
| Capability | Status | Notes |
|---|---|---|
| Encryption at rest | Live | Rule and context bodies are ciphertext; only frontmatter is plaintext. AES-256-GCM for packs, SOPS+age for the context repo. (C1) |
| Decryption in process memory only | Live | No plaintext is written to disk. Enforced as a hard constraint in taip-agent. (C2) |
| Role filtering before decryption | In alpha | Frontmatter RBAC runs per call and fails closed. The identity behind the role is asserted, not authenticated — a caller can claim a different profile. (C3) |
| Authenticated origin auth and licence enforcement | Planned | The mechanism exists in code but is off in production deploys. Only Cloudflare Access protects tenants today. (C4) |
| EU tenant isolation | Live | Each tenant runs in its own container with its own encrypted repository and keys, hosted in the EU. (C5) |
| Audit log — context-mcp | Live | Every read and write through context-mcp is logged with actor, role, item, and timestamp. (C6) |
| Audit log — brain | Planned | Brain-side audit logging is not yet built. This is a beta blocker. (C7) |
| Append-only audit log semantics | Planned | Not evidenced for brain; unverified for context-mcp. Nothing claims append-only until C7 ships and the property is documented. (C8) |
| Crypto-shred for GDPR Art. 17 | In alpha | Per-subject crypto-shred exists. No verified-restore or deletion runbook yet; backup verification is a beta gate. (C9) |
| Tenant-level teardown | Planned | No automated teardown of containers, DNS, Access policies, secrets, or backups. (C10) |
| Key revocation ends access everywhere | Planned | Validation is offline-JWT only. A leaked licence is valid until expiry. Revoking the key ends access to TAIP-managed content once the licence expires. (C11) |
| Same rules across every MCP client | Live | Cursor, Copilot, JetBrains AI, Claude — any client that speaks MCP reads the same rules. (C13) |
| MCP tool catalogue | Live | A curated catalogue of MCP tools, published to your assistants via the taip-tools proxy. (C14) |
| Policy enforcement on tool execution | Planned | Out of scope for the current product. taip serves its own content and is not a gateway or firewall over other tools. (C15) |
Status as at 2026-07-29. Every row traces to claims-register.md §3 (docs/copy-sources/claims-register.md).
What taip is not.
Not a gateway or firewall. taip serves its own content to AI assistants. Your other tools work exactly as before.
Not real-time enforcement over every tool call. Policy enforcement on tool execution is out of scope today (C15, planned). taip governs what assistants read from taip — not what they do elsewhere.
Not instant key revocation. Destroying a key ends access to TAIP-managed content, but licence validation is offline-JWT only. A leaked licence remains valid until expiry (C11, planned).
Brain-side audit logging is not live. context-mcp logs every read and write (C6, live). brain does not yet (C7, planned). Do not rely on brain audit logs for compliance evidence at this stage.
enterprise pilot
Request a pilot.
We onboard companies one at a time. No spam. Unsubscribe any time.