Truveil

Install the SDK

Add Truveil to your project in seconds. Available for JavaScript and Python — no external dependencies.

Log your agent actions

Add one line at each decision point in your agent. Every autonomous action gets captured automatically.

Get your audit report

Click Analyse — get a plain-language compliance report instantly. Export as PDF for regulators.

const Truveil = require('./truveil');
const truveil = new Truveil(
  'https://api.truveil.app',
  'YOUR_API_KEY_ABOVE',
  'My Agent'
);

const run = truveil.run();
run.decision('subject-001', {
  factors: ['criterion met', 'risk within threshold'],
  outcome: 'approve',
  confidence: 0.9,
});
run.stage('screening').complete('1 of 1 processed');

from truveil_sdk import TruveilSDK

sdk = TruveilSDK(
    'https://api.truveil.app',
    'YOUR_API_KEY_ABOVE',
    'My Agent'
)

run = sdk.run()
run.decision('subject-001',
    factors=['criterion met', 'risk within threshold'],
    outcome='approve',
    confidence=0.9)
run.stage('screening').complete('1 of 1 processed')

Active agents

—

of — tracked

High risk

—

needs review

Runs this week

—

across the fleet

As of

—

auto-refreshes hourly

Needs attention

checking…

All agents

Loading agents…

LIVE AGENT ACTIVITY

Waiting for your agent. Select a project above or connect your agent using the SDK.

or explore demo scenarios

Select a run to see the audit preview

Manual Analysis

Connect your AI agent

● Your API Key

Loading…

Use this key in any of the integrations below. Copy works without revealing it.

Closing ends your access now, and your API key stops working immediately. Your remaining personal account data is de-identified 30 days after closure. Your AI agent decision logs are audit records: they are not deleted on closure and are retained under applicable data-retention requirements. You can request erasure of specific logs at any time.

Claude, ChatGPT, or any MCP client

Connect Truveil as a tool. No code.

Standalone Python or JavaScript agent

Custom AI agent you have built and run yourself.

ChatGPT Workspace Agent

Agent built inside a ChatGPT Business or Enterprise workspace.

Custom GPT

GPT you have built in ChatGPT.

n8n workflow

Dedicated community node. Credential-based.

Click any platform above to see the step-by-step setup.

Connect Claude, ChatGPT, or any MCP client

What you need: An MCP-capable client. Some clients sign you in directly and need no key at all.

Option A — Sign in directly (recommended)

If your client supports connector sign-in, this is the shortest path. No key to copy, no header to configure.

01Add a custom connector using the URL below.
02Your client opens a Truveil sign-in page. Sign in and approve access.
03In any conversation, say: “Use Truveil”.

https://mcp.truveil.app/mcp

Option B — API key header

For clients that ask for a custom header instead of signing you in. Same URL as above.

01Add a custom MCP server using the connector URL above.
02Set the authentication header shown below.
03In any conversation, say: “Use Truveil”.

x-client-secret: ••••••••••••••••

That is a live credential. Treat it like a password: never paste it into a shared document, a ticket, or a public repository.

Test it Ask your assistant to log a decision. Open Live Feed in your dashboard and watch it arrive.

Connect your standalone agent

What you need: A Truveil API key from your dashboard, plus Python 3.7+ or Node.js 16+.

01Download the Truveil SDK file: truveil_sdk.py (Python) or truveil.js (JavaScript). Drop it into your project folder.
02Initialise the SDK once at the top of your agent file using your API key.
03Open a run handle with sdk.run() and record each decision with one named line, as below.

        
        
      
from truveil_sdk import TruveilSDK

sdk = TruveilSDK(
    'https://api.truveil.app',
    'your_truveil_api_key',
    'My Agent'
)

run = sdk.run()
run.decision('subject-001',
    factors=['criterion met', 'risk within threshold'],
    outcome='approve',
    confidence=0.9)
run.stage('screening').complete('1 of 1 processed')
const Truveil = require('./truveil');

const truveil = new Truveil(
  'https://api.truveil.app',
  'your_truveil_api_key',
  'My Agent'
);

const run = truveil.run();
run.decision('subject-001', {
  factors: ['criterion met', 'risk within threshold'],
  outcome: 'approve',
  confidence: 0.9,
});
run.stage('screening').complete('1 of 1 processed');

Test it Run your agent. Events appear in your Live Feed within seconds.

Connect your Workspace Agent

What you need: A ChatGPT Business, Enterprise, Edu, or Teachers plan with Developer Mode enabled, plus your Truveil API key.

01In ChatGPT, go to Workspace settings → Apps → enable Developer Mode.
02Open your Workspace Agent → Skills → Add custom MCP server.
03Paste the URL: https://mcp.truveil.app/mcp
04Set the header x-client-secret to your Truveil API key.
05Add the integration paragraph below to your agent's instructions.

You have access to the Truveil audit tool. After every material
decision, action, or recommendation you make, call the appropriate
Truveil tool before proceeding to the next step. Set project to the
name of this agent, run_id to a short identifier for this session,
and agent to the same name as the project. Include your reasoning in
the detail field. Do not assess or assign risk levels yourself.
Do not combine multiple decisions into one log entry. One decision,
one log entry. On completion, call get_audit_history with the project
name and report the total events logged.

Test it Run your agent on a task. Open Live Feed in your dashboard and watch events arrive.

Connect your Custom GPT

What you need: A ChatGPT Plus, Pro, Team, Business, or Enterprise plan, plus your Truveil API key.

01Open your Custom GPT → Configure → Actions → Create new action.
02Paste the OpenAPI schema below.
03Set Authentication to API Key, Custom header x-client-secret, value: your Truveil API key.
04Add the integration paragraph below to your GPT's Instructions.

openapi: 3.1.0
info:
  title: Truveil Audit
  version: 1.1.0
servers:
  - url: https://api.truveil.app
components:
  securitySchemes:
    apiKeyAuth:
      type: apiKey
      in: header
      name: x-client-secret
security:
  - apiKeyAuth: []
paths:
  /log:
    post:
      operationId: logAgentEvent
      summary: Log a single agent decision or action
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              required: [action, detail, project]
              properties:
                project: { type: string }
                run_id: { type: string }
                step: { type: integer, description: "A sequential integer marking this event's position in the run." }
                agent: { type: string }
                action: { type: string }
                detail: { type: string }
                risk: { type: string, description: "Risk level for this event: LOW, MEDIUM or HIGH. Omit it if you have not assessed one; the row then records no level, rather than one Truveil chose." }
                decision_type: { type: string }
                human_checkpoint: { type: boolean }
                reversible: { type: boolean }
                checkpoint_ref: { type: string, description: "Reference to a preceding checkpoint: the approval_id returned by a gate_confirmed response (request_approval, request_consequential_decision, or POST /approvals/request-consequential). Recorded verbatim; the write boundary checks only that it is a string, and whether it names a real issued gate id is scored at report time, not enforced here." }
                event_type: { type: string, enum: [output, synthesis, ingestion, decision, disclosure, correction], description: "What kind of event this row records, from the closed set output|synthesis|ingestion|decision|disclosure|correction. Optional; anything outside the set is rejected by name. This is the caller's own classification of the event, never defaulted." }
      responses:
        '200': { description: Event logged }
  /logs:
    get:
      operationId: getAuditHistory
      summary: Get audit history for a project
      parameters:
        - name: project
          in: query
          required: true
          schema: { type: string }
      responses:
        '200': { description: Logs returned }

You have access to the Truveil audit tool. After every material
decision, action, or recommendation you make, call logAgentEvent
before proceeding to the next step. Set project to the name of this
assistant, run_id to a short identifier for this session, and agent
to the same name as the project. Include your reasoning in the detail
field. Do not log secrets, credentials, or personal data in
detail fields.

Test it Ask your GPT to run a task. Open Live Feed in your dashboard and watch events arrive.

Connect your n8n workflow

What you need: An n8n instance where you can install community nodes, plus your Truveil API key.

01In n8n: Settings → Community Nodes → Install, then enter the package name below.
02Create a Truveil API credential and paste your API key into it. Copy it from the card at the top of this panel.
03Add Truveil steps to any workflow. Registration, decision logging, approvals, and audit history are all node operations.

n8n-nodes-truveil

This is a purpose-built node, not an MCP attachment. Your credential is stored in n8n’s credential vault and sent as a header, never in a URL.

Test it Run the workflow once. Open Live Feed in your dashboard and watch events arrive.