Cleanlab

Cleanlab is a data-and-AI trust platform. Its Trustworthy Language Model (TLM) wraps any LLM with a real-time trustworthiness score to catch hallucinations, Cleanlab Studio curates and labels training data and deploys reliable ML models, and Codex adds a safety, guardrail, and remediation layer for AI assistants and RAG. The open-source cleanlab library underpins it all with data-centric AI.

4 APIs 0 Features

AILLMTrustworthinessData QualityGuardrails

APIs

Trustworthy Language Model (TLM)

Wraps any base LLM (GPT, Claude, Gemini, Llama, and more) with a real-time 0-1 trustworthiness score to detect hallucinations and unreliable answers. Exposed via an OpenAI-compa...

Cleanlab Studio

Cloud platform for data-centric AI - upload datasets, run automated data curation and error detection, train AutoML models, and deploy them for real-time REST inference. Driven ...

Codex

Safety and reliability layer for AI assistants and RAG. Validates every AI response, decides whether to guardrail it, scores it across multiple eval criteria, escalates to subje...

Open-Source Library

The original open-source cleanlab Python library for data-centric AI - finds and fixes label errors, outliers, and other data issues using confident learning. Runs locally (pip ...

Collections

Cleanlab API

OPEN

Pricing Plans

Cleanlab Plans Pricing

5 plans

PLANS

Rate Limits

Cleanlab Rate Limits

4 limits

RATE LIMITS

FinOps

Cleanlab Finops

FINOPS

Event Specifications

Cleanlab TLM Chat Completions Streaming (HTTP + SSE)

AsyncAPI 2.6 description of the Trustworthy Language Model (TLM) **chat completion streaming** surface. Cleanlab does not publish a WebSocket API. The TLM is exposed via an Open...

ASYNCAPI

Resources

👥

GitHubOrganization

Sources

opencollection: 1.0.0
info:
  name: Cleanlab API
  version: '1.0'
request:
  auth:
    type: bearer
    token: '{{bearerToken}}'
items:
- info:
    name: TLM
    type: folder
  items:
  - info:
      name: Trustworthy chat completion (OpenAI-compatible)
      type: http
    http:
      method: POST
      url: https://api.cleanlab.ai/api/v1/openai_trustworthy_llm/v1/chat/completions
      body:
        type: json
        data: "{\n  \"model\": \"gpt-4.1-mini\",\n  \"messages\": [\n    { \"role\": \"user\", \"content\": \"What is the\
          \ capital of France?\" }\n  ],\n  \"stream\": false\n}"
    docs: OpenAI-compatible chat completion returning the completion plus a trustworthiness score in tlm_metadata. Set stream=true
      for an SSE stream.
- info:
    name: Codex
    type: folder
  items:
  - info:
      name: Validate an AI response
      type: http
    http:
      method: POST
      url: https://api.cleanlab.ai/api/v1/codex/projects/{project_id}/validate
      body:
        type: json
        data: "{\n  \"query\": \"What is your refund policy?\",\n  \"context\": \"Retrieved context passed to the LLM.\",\n\
          \  \"response\": \"The AI-generated answer being validated.\"\n}"
    docs: Validates a response and returns should_guardrail, eval_scores, escalated_to_sme, expert_answer, and log_id. Authenticate
      with the project access key.
- info:
    name: Studio
    type: folder
  items:
  - info:
      name: Predict with a deployed model
      type: http
    http:
      method: POST
      url: https://api.cleanlab.ai/api/v1/deployment/{model_id}/predict
      body:
        type: json
        data: "{\n  \"data\": [\n    \"first record to classify\",\n    \"second record to classify\"\n  ],\n  \"return_pred_proba\"\
          : true\n}"
    docs: Runs inference against a deployed Cleanlab Studio model and returns predicted labels (and optional probabilities).