# Rasepi

> Rasepi is a trust-control plane for company knowledge. Before an AI assistant, RAG pipeline, agent, or person relies on a document, Rasepi answers one question: can this knowledge be safely used right now, by this actor, for this purpose?

The flagship is the **Rasepi Trust Engine** (powered by the Trust Gate API): it evaluates each candidate source and returns allow / warn / block / escalate, with reasons and obligations, so only trusted knowledge reaches the LLM. The core principle: do not only guard the generated answer — guard the knowledge before it enters the prompt. Rasepi is not an LLM proxy; it never reads your prompts and requires no migration.

The site is available in German (root) and English (/en/).

## Products

- **Rasepi Trust Engine** (the landing page, https://rasepi.com): the AI-trust layer. Gate what your AI is allowed to use; allow/warn/block decisions with reasons; connect any source; full audit ledger. Machine access via the Trust API and MCP.
- **Rasepi Docs** (https://rasepi.com/produkt/docs.html): the knowledge platform — per-document trust/freshness score, block-level translation (40+ languages), reviews and expiry, real-time collaboration, reader portal, ask-your-docs.
- **Rasepi Capture** (https://rasepi.com/produkt/capture.html): voice capture that turns speech into structured documents in 40+ languages.

## What Rasepi is not

- Not a generic LLM proxy (does not sit between you and your model, does not read prompts).
- Not an output-moderation/guardrails product (the problem is untrustworthy source knowledge, not toxic output).
- Not a simple freshness scanner (freshness is one input; the decision is contextual trust — actor, action, channel).

## Trust decision model

Decisions: ALLOW, WARN, BLOCK, ESCALATE. Each carries machine-readable reasons (code, dimension, severity, message), a freshness score, classification, policy source, and obligations (citation required, disclaimer required). Dimensions: freshness, ownership, provenance, policy, sensitivity, request context. The same document can be allowed for an internal human and blocked for a customer-facing bot or an action-taking agent.

## Pages

- [Home — Trust Engine](https://rasepi.com/)
- [Rasepi Docs](https://rasepi.com/produkt/docs.html)
- [Rasepi Capture](https://rasepi.com/produkt/capture.html)
- Use cases: [Secure AI answers](https://rasepi.com/loesungen/ki-absichern.html), [Support bots](https://rasepi.com/loesungen/support-bots.html), [Stale knowledge](https://rasepi.com/loesungen/veraltete-doku.html), [Compliance & audit](https://rasepi.com/loesungen/compliance-audit.html)
- Audiences: [AI & platform teams](https://rasepi.com/fuer/ki-plattform-teams.html), [Compliance & governance](https://rasepi.com/fuer/compliance-governance.html), [Docs & knowledge teams](https://rasepi.com/fuer/doku-wissensteams.html)
- [Request a demo](https://rasepi.com/demo.html)
- [Open Source](https://rasepi.com/open-source.html)
- [Contact](https://rasepi.com/contact.html) · [Imprint](https://rasepi.com/imprint.html) · [Privacy](https://rasepi.com/privacy.html)

## Developer resources

- [Developer portal](https://developers.rasepi.com)
- [Getting started](https://developers.rasepi.com/getting-started)
- [Trust API reference](https://developers.rasepi.com/trust)
- [REST API reference](https://developers.rasepi.com/api)
- [MCP server](https://developers.rasepi.com/mcp)
- [Developer llms.txt](https://developers.rasepi.com/llms.txt)

## Trust API (machine surface)

- Base URL: `https://api.rasepi.com`; auth via tenant API keys (`Bearer rk_…` / `X-Api-Key`) or OAuth client credentials; scopes `trust:evaluate`, `trust:read`, `trust:usage`.
- `POST /api/trust/evaluate` — evaluate up to 100 sources for an actor/action/channel.
- `POST /api/trust/filter-sources` — RAG post-retrieval filter; returns allowed / warned / blocked.
- `GET /api/trust/sources/{sourceId}` — decision-free trust metadata for badges and dashboards.
- `POST /api/trust/usage` — log that an AI system relied on certain sources.
- `GET /api/trust/decisions/{decisionId}` — full audit record of a past decision.
- Source ID forms: `entry:{guid}`, `{connectorKey}:{externalId}` (e.g. `confluence:SPACE:12345`), or http(s) URLs.

## Tech stack

- Backend: .NET 8, ASP.NET Core, EF Core, SignalR, PostgreSQL + pgvector
- Frontend: Vue 3, TypeScript, Vuetify 3, TipTap, Pinia
- Auth: OpenIddict (OAuth 2.0 / OIDC), tenant API keys for machine access
- AI: OpenAI-compatible providers, semantic search, agent conversations
- Deployment: Docker, Cloudflare Pages