Developers

From zero to governed in under an hour.

A versioned REST API (/api/v1/), native MCP support, and token-based auth. Async training via Celery + Redis. pgvector-native knowledge base.

POST · /API/V1/AI/INFER

Governed LLM inference

Runs input filter + classifiers, calls the configuration's model, screens output. Returns flags and the six validation scores.

POST /api/v1/ai/infer
Authorization: Token <key>

{
  "config_id": "<uuid>",
  "messages": [
    {"role":"user","content":"Summarize Q3..."}
  ],
  "image_urls": []
}

→  { "message_id": "...",
      "content": "...",
      "flags": { "attack": false,
                 "impersonation": false,
                 "disallowed": false,
                 "sensitive": false },
      "validation_scores": {
        "religion":0.01, "politics":0.04,
        "gender":0.00, "race":0.01,
        "violence":0.02, "sensitive":0.06 },
      "tokens_used": { "prompt":120,"completion":88 } }

KNOWLEDGE BASE

Search & train DocumentCollections

Semantic + full-text hybrid retrieval over pgvector. Training is async — poll a job ID or subscribe via webhook.

POST /api/v1/kb/search
{ "collection_id": "<uuid>",
  "query": "revenue recognition policy",
  "top_k": 5 }

POST /api/v1/kb/train
{ "collection_id": "<uuid>",
  "document_id":   "<uuid>" }

→  { "job_id": "<uuid>" }

GET  /api/v1/jobs/<job_id>/status
→  { "status": "succeeded" }

MODERATION · AS-A-SERVICE

Moderate or redact arbitrary text

Standalone endpoints. Use before storing user-generated content downstream, or as a governance pass on legacy pipelines.

POST /api/v1/moderation/moderate
{ "text": "..." }

POST /api/v1/moderation/redact
{ "text": "..." }

→  { "flags": { "sensitive": true },
      "scores": { "religion":0.02, ... },
      "redacted": "Please contact me at [EMAIL]." }

MCP · SERVICE ACCOUNTS

Governed tool calls from external MCP servers

Org-scoped service accounts. Tokens shown once, hashed at rest, instant-revoke. Every MCP-initiated message is filtered, scored, and audit-logged.

// External MCP server calls 2Trust on org's behalf:
fetch("/api/v1/ai/infer", {
  headers: {
    "Authorization": "Token mcp_sa_4f2a...",
    "Content-Type":  "application/json"
  },
  body: JSON.stringify({
    config_id: "<uuid>",
    messages:  [...],
    meta:      { mcp_tool: "crm.sync" }
  })
});

Quickstart/api/v1/ai/infer/api/v1/kb/search/api/v1/kb/train/api/v1/moderation/*/api/v1/jobs/*MCP service accountsWebhooksAuth0 SSOSelf-hosted installDocker + CodeDeploypgvector setup

Stack

Open, inspectable, deployable in your VPC.

Backend

Django 4.2 + DRF · Python 3.10+

Data

PostgreSQL + pgvector

Async

Redis + Celery 5.3

PyTorch 2.1 · HF Transformers · FP16/INT8

Frontend

Django templates · Tailwind 3.4

Deploy

Docker · CodePipeline · CodeDeploy · ECR

Identity

Auth0 OAuth2 · Django Allauth

Observability

ErrorLog · simple-history · perf dash