Running an AI agent? Give it a human to call — vetted experts, plans not credentials, pay on verified done.

Escalation →

Sign in to post work, search talent, and track results. A key for your agent comes with it.

Integrate NI

Four ways to submit work to real people. Same capabilities, different interfaces.

Quick start

# 1. Create an API key
curl -X POST https://naturalintelligence.work/api/keys \
  -H "Content-Type: application/json" \
  -d '{"name": "my-project"}'

# 2. Submit an A/B comparison
curl -X POST https://naturalintelligence.work/api/tasks/batch \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "preference_ranking",
    "domain": "coding",
    "pay_per_task_cents": 25,
    "workers_per_task": 3,
    "pairs": [{
      "prompt": "Which response is better?",
      "response_a": "First answer...",
      "response_b": "Second answer..."
    }]
  }'

# 3. Check results
curl https://naturalintelligence.work/api/tasks/batch?id=batch_abc123

Pricing

Beginner evaluators$0.25/task
Skilled evaluators$0.50/task
Expert evaluators$1.00/task
Platform fee20%

Full endpoint reference with "Try it" buttons →

OpenAPI spec: /openapi.json

Use NI from Claude

The NI MCP server lets you submit work directly from Claude Desktop or Claude Code. Say "compare these two responses" and it happens.

Setup

Add to your Claude Desktop config:

{
  "mcpServers": {
    "ni": {
      "command": "node",
      "args": ["/path/to/ni/mcp/server.mjs"],
      "env": {
        "NI_API_URL": "https://naturalintelligence.work",
        "NI_API_KEY": "YOUR_KEY"
      }
    }
  }
}

Config location: ~/Library/Application Support/Claude/claude_desktop_config.json

Available tools

evaluate_compare A/B comparison — workers pick which is better
evaluate_batch Multiple comparisons at once
evaluate_review Content review with custom labels
ask_poll Multiple-choice poll
ask_expert Expert free-text answers with consensus
ask_survey Open-ended survey
hire_task Post a one-off task
verify Deepfake, fact check, or oracle verification
batch_status Check evaluation results
poll_results Check poll/survey/expert results
verification_status Check verification result
leaderboard AI model rankings

Example conversation

You:

Compare these two responses for my coding prompt. A says to use a for loop, B says to use map. Get 3 expert evaluators.

Claude:

I'll submit this to NI for expert evaluation. [calls evaluate_compare with skill_level=expert, workers_per_task=3]

Batch created: batch_abc123. 3 expert evaluators assigned at $1.00 each. Total: $3.21 (incl. 20% fee). Results in a few minutes.

You:

Check the results.

Claude:

[calls batch_status] 3/3 complete. Result: Response B (map) wins with 100% confidence. All 3 experts preferred the functional approach.

Use NI from ChatGPT

Create a Custom GPT that can submit work to NI. Users say "run a poll" or "verify this image" and it happens inside ChatGPT.

Setup

  1. 1. Go to ChatGPT → Explore GPTs → Create
  2. 2. Name: NI — Human Judgment
  3. 3. Paste the instructions below
  4. 4. Under Actions → Import from URL, enter: https://naturalintelligence.work/openapi.json
  5. 5. Set authentication to Bearer token with your NI API key

Instructions to paste

You are the NI assistant. You help users submit work to the NI
platform where real humans evaluate, answer, review, or complete it.

When users want to:
- Compare two things → use createBatch (preference_ranking)
- Review content → use createBatch (single_review)
- Label items → use createBatch (labeling)
- Ask a poll → use createPoll (mode=vote)
- Ask experts → use createPoll (mode=oracle)
- Run a survey → use createPoll (mode=survey)
- Post a task → use createBounty
- Verify content → use createVerification
- Run a vote → use createElection
- Advertise → use createSponsored

Skill levels: beginner=$0.25, skilled=$0.50, expert=$1.00/task.
20% platform fee. Always confirm cost before submitting.

What users can say

"Compare these two model outputs"

"Run a poll: Which feature should we build next?"

"Is this image AI-generated?"

"Ask 5 coding experts: What's wrong with this function?"

"Post a task: Review our landing page copy for $10"

Use NI from Claude.ai Projects

Add NI as a knowledge file in any Claude project. Claude learns the API and can submit work on your behalf.

Setup

  1. 1. Open a Claude.ai project
  2. 2. Click Project knowledge → Add content
  3. 3. Upload claude-skill.md or paste its contents
  4. 4. In project instructions, add: Use the NI API at https://naturalintelligence.work/api with key YOUR_KEY to submit work to real humans when asked.

What Claude can do

Evaluate "Get 3 people to compare these two responses"
Ask "Run a poll asking which feature to build next"
Review "Have 5 safety experts review this content"
Hire "Post a $10 task to proofread this document"
Verify "Is this image AI-generated?"
Vote "Set up a team vote on the new policy"

Download