unformal-api（unformal API 搜索）

Unformal API

STAY CURRENT — check before creating a Pulse

Before you create or modify a Pulse, make sure your local CLI and this skill file are up to date. The Pulse API gains new config fields regularly (persona, welcome copy, research toggle, topics, etc.) and an outdated CLI / skill silently drops them — the Pulse gets created with defaults the user didn't intend.

One curl tells you everything:

curl -fsS https://unformal.ai/api/v1/version

Returns the current CLI version, current skill version, and the minimum versions that know about the full API surface. Then:

1. Compare the cli.latest to unformal --version. If yours is older, run npm i -g @unformal/cli@latest.
2. Compare the skill.latest to the version: in the frontmatter of this file (currently 1.4.0). If yours is older, re-install: mkdir -p ~/.claude/skills/unformal && curl -fsS https://unformal.ai/SKILL.md > ~/.claude/skills/unformal/SKILL.md, then read the new version before continuing.
3. If you hit 409 / 400 / missing-field errors from the API, the first thing to check is always: are you on the latest CLI + skill?

Skipping this step is the #1 cause of "the Pulse I created is missing fields I wanted." Don't skip it.

Create AI-powered conversational flows (Pulses) that replace forms, surveys, and intake emails. Send someone a link — an AI agent has a real conversation with them and extracts structured data.

When to use this skill

Invoke this skill whenever the user wants to collect information from other people through a link (as opposed to talking to the user directly). Concrete triggers:

• Feedback / NPS / surveys — "gather feedback on X", "send out a survey", "ask my customers how they feel about Y"
• User interviews / customer research — "I want to interview 20 users about Z", "run a research sprint on pain points"
• Lead qualification — "qualify inbound leads", "build a discovery flow for new prospects"
• Intake / onboarding — "collect requirements before a kickoff call", "new-client intake", "application form for our program"
• Event / workshop / community check-ins — "collect reflections from attendees", "post-event feedback"
• Beta / product testing feedback — "get structured feedback from beta users"
• HR / hiring / applications — "screening questions for candidates", "employee pulse check"
• Anything where a form feels too static — the user wants follow-up questions, clarifications, or open-ended answers turned into structured fields

If the user says "I want to send this to [N] people and get structured answers back," Unformal is the right tool. If they want to talk to the user directly in the current chat, it is not.

Typical flow

1. Ask clarifying questions: what are they trying to learn, who will receive the link, tone (casual / professional), how many questions max, what structured fields to extract.
2. Create the Pulse with POST /api/v1/pulses (or unformal create via the CLI).
3. Give the user the shareable https://unformal.ai/p/<slug> link.
4. Once responses come in: use unformal conversations <id>, unformal resonance <id>, or the equivalent API endpoints to read individual transcripts and aggregate insights.

Setup (existing account)

If the user already has a verified Unformal account but doesn't have their API key saved, use the login flow to mint a fresh one. API keys are only stored hashed, so a previously-issued key can never be recovered.

curl -X POST "https://unformal.ai/api/v1/login" \
  -H "Content-Type: application/json" \
  -d '{"email": "your@email.com"}'

# Check inbox, then:
curl -X POST "https://unformal.ai/api/v1/login/verify" \
  -H "Content-Type: application/json" \
  -d '{"email": "your@email.com", "code": "123456"}'
# → {"data": {"api_key": "unf_...", "workspace_id": "...", "status": "authenticated"}}

Via the CLI: unformal login --email your@email.com (emails code), then unformal login --email your@email.com --code 123456 --save (verifies + saves to ~/.unformal/config).

Setup (new account)

1. Sign up via API (no browser needed):

curl -X POST "https://unformal.ai/api/v1/signup" \
  -H "Content-Type: application/json" \
  -d '{"email": "your@email.com"}'

1. Check your email for a 6-digit verification code
2. Verify:

curl -X POST "https://unformal.ai/api/v1/verify" \
  -H "Content-Type: application/json" \
  -d '{"email": "your@email.com", "code": "123456"}'

1. Your API key (from step 1) is now active

Or: visit unformal.ai/studio/settings to create an API key manually.

Base URL: https://unformal.ai/api/v1

Auth: Authorization: Bearer unf_YOUR_KEY on every request.

Quick Start

Create a Pulse and get a shareable link

curl -X POST "https://unformal.ai/api/v1/pulses" \
  -H "Authorization: Bearer unf_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"intention": "Understand what a new client needs before our first meeting"}'

Response:

{
  "data": {
    "id": "pls_abc123",
    "url": "https://unformal.ai/p/your-slug",
    "slug": "your-slug",
    "status": "active"
  }
}

Send the URL to anyone. The AI conducts the conversation. You get structured data back.

Endpoints

Create Pulse — Request Body

Modes

Interview (default): Guided conversation. The AI asks one question at a time, adapting follow-ups. Best for discovery, qualification, deep interviews.

Extract: Brain dump first. The respondent writes everything upfront. The AI asks only about gaps. Best for intake forms, bug reports, data collection.

Example: Minimal

{"intention": "Qualify this lead for our enterprise plan"}

Example: Full

{
  "intention": "Understand what motivates each team member and where they need support",
  "context": "We're a 15-person startup. This is our quarterly team check-in.",
  "persona": "You are a warm, empathetic coach who asks thoughtful follow-ups. Use first names. Be encouraging but probe deeper when someone gives a surface-level answer.",
  "mode": "interview",
  "tone": "coaching",
  "maxQuestions": 10,
  "completionPriority": "coverage",
  "outputFields": [
    {"name": "motivation", "type": "string", "description": "What currently motivates this person"},
    {"name": "blockers", "type": "array", "description": "Things preventing them from doing their best work"},
    {"name": "support_needed", "type": "string", "description": "What support they need from leadership"},
    {"name": "energy_level", "type": "number", "description": "Self-reported energy/morale 1-10"}
  ],
  "showInsights": true,
  "webhookUrl": "https://your-app.com/webhooks/unformal"
}

Example: Extract mode with persona

{
  "intention": "Screen startup founders applying for pre-seed investment",
  "persona": "You are a sharp, funny VC intern who has seen 10,000 bad pitch decks. Roast gently, find the diamonds.",
  "mode": "extract",
  "tone": "custom",
  "maxQuestions": 10,
  "allowResearch": true,
  "completionPriority": "coverage",
  "outputFields": [
    {"name": "company_name", "type": "string", "description": "Startup name"},
    {"name": "one_liner", "type": "string", "description": "One-sentence description"},
    {"name": "traction", "type": "string", "description": "Users, revenue, or proof of demand"},
    {"name": "pitch_score", "type": "number", "description": "AI-assigned score 1-10"}
  ]
}

Update Pulse — PATCH /pulses/:id

All creation fields are patchable. Send only the fields you want to change.

curl -X PATCH "https://unformal.ai/api/v1/pulses/PULSE_ID" \
  -H "Authorization: Bearer unf_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"maxQuestions": 12, "outputFields": [{"name": "score", "type": "number", "description": "Lead score"}]}'

Retrieving raw responses

Three ways to pull individual respondents' answers.

1. List with echoes (fast, no transcripts)

curl "https://unformal.ai/api/v1/pulses/PULSE_ID/conversations" \
  -H "Authorization: Bearer unf_YOUR_KEY"

Returns [{id, status, echo, metadata, completedAt, createdAt}]. Good for scanning who completed and what structured fields were extracted. No transcripts.

2. Single conversation with full transcript

curl "https://unformal.ai/api/v1/conversations/CONV_ID" \
  -H "Authorization: Bearer unf_YOUR_KEY"

Returns everything: transcript (every AI message + user reply, including responseData for structured answers like slider values, multi-select picks, ranking orders) and echo (extracted structured fields + key quotes + summary).

3. Bulk export — everything for all conversations

curl "https://unformal.ai/api/v1/pulses/PULSE_ID/export?format=json" \
  -H "Authorization: Bearer unf_YOUR_KEY" \
  -o all-responses.json

Returns {pulse, conversations: [...with transcript and echo...], exportedAt, total}. Use format=csv for a flat tabular dump of echo fields.

Transcript shape

Each transcript message has:

{
  "role": "user" | "assistant",
  "content": "...",
  "timestamp": 1776343572188,
  "uiHint": { "type": "slider", "config": {...} },     // on assistant msgs with structured questions
  "responseData": { "type": "slider", "data": {"value": 8} }  // on user msgs that answered a structured question
}

responseData.type matches the ui_hint type: slider, quick_options, multi_select, ranking, image_select. The data shape:

• slider → {value: number}
• quick_options → {selected: string} (single)
• multi_select / image_select → {selected: string[]}
• ranking → {order: string[]} (first element = most valued)

Echo — Structured Output

When a conversation completes, the AI extracts structured data:

{
  "echo": {
    "fields": {
      "budget_range": "$50k-100k",
      "timeline": "Q3 2026",
      "current_tools": ["Salesforce", "HubSpot"]
    },
    "summary": "Strong fit. Budget aligned, timeline Q3.",
    "keyQuotes": ["We spend 3 hours daily on data entry"],
    "subtext": "Enthusiastic but hesitant about buy-in.",
    "sentimentScore": 7
  }
}

Resonance — Aggregate Insights

Once a Pulse has 3+ completed conversations, Unformal synthesizes an aggregate view:

curl "https://unformal.ai/api/v1/pulses/PULSE_ID/resonance" \
  -H "Authorization: Bearer unf_YOUR_KEY"

Response (abridged):

{
  "data": {
    "available": true,
    "totalConversations": 14,
    "autoSummary": "WAT has successfully built a highly valued founder community...",
    "themes": [{"theme": "Community value", "frequency": 12, "sentiment": "positive"}],
    "consensusPoints": ["Free coworking space is the most valued beta offering"],
    "divergencePoints": ["Views split on event format — founder stories vs workshops"],
    "recommendedActions": ["Hire a community manager", "Add call booths"],
    "sentimentDistribution": {"positive": 64, "neutral": 29, "negative": 7},
    "questionAggregates": [
      {
        "questionIndex": 10,
        "question": "How likely are you to recommend WAT?",
        "type": "slider",
        "stats": {
          "n": 14, "mean": 7.9, "median": 8, "min": 5, "max": 10,
          "distribution": [{"bucket": "1-2", "count": 0}, ...],
          "nps": {"score": 14, "promoters": 5, "passives": 6, "detractors": 3}
        }
      },
      {
        "questionIndex": 3,
        "question": "Which WAT channels are you part of?",
        "type": "multi_select",
        "stats": {"n": 10, "options": [{"label": "WAT Connect", "count": 10, "percentage": 100}, ...]}
      },
      {
        "questionIndex": 4,
        "question": "Rank these offerings from most to least valuable:",
        "type": "ranking",
        "stats": {"n": 12, "items": [{"label": "Free coworking space", "avgRank": 1.92, "topCount": 5}, ...]}
      }
    ],
    "openEndedHighlights": [
      {
        "fieldName": "pain_point",
        "label": "Pain Point",
        "description": "The single thing they wish were 10x better",
        "values": ["Community management gap", "More call booths", ...]
      }
    ],
    "featuredQuotes": ["a community is built around people, if they're not connecting it's not working", ...]
  }
}

If < 3 conversations completed, returns {"available": false, "reason": "..."}.

Question aggregate shapes

• slider — {n, mean, median, min, max, distribution[], nps?}. NPS is auto-detected when a 1-10 (or 0-10) slider question contains "recommend" or similar.
• ranking — {n, items: [{label, avgRank, topCount, rankCounts[]}]}. Lower avgRank = more valued.
• multi_select / quick_options — {n, options: [{label, count, percentage, shareOfPicks}]}
• image_select — same as multi_select, plus images: [{label, url}]

Analytics — Completion and Engagement

curl "https://unformal.ai/api/v1/pulses/PULSE_ID/analytics" \
  -H "Authorization: Bearer unf_YOUR_KEY"

Returns completionRate, avgDuration, avgFieldCoverage (how much of the structured output was filled on average), avgSentiment, avgMessageCount, avgResponseLength, per-field fieldFillRates, abandonmentRate and commonAbandonmentPoint (which question tends to cause drop-off).

Webhooks

Set webhookUrl on a Pulse to receive Echo data on completion:

{
  "event": "conversation.completed",
  "conversationId": "conv_123",
  "pulseId": "pls_456",
  "echo": {"fields": {...}, "summary": "...", "keyQuotes": [...], "sentimentScore": 7},
  "completedAt": "2026-03-29T10:00:00Z"
}

Signed with X-Unformal-Signature (HMAC-SHA256).

Common Patterns

Agent creates intake for each new lead

1. POST /pulses with intention for the specific lead
2. Send the URL to the lead
3. Wait for webhook with Echo data
4. Update CRM with structured qualification data

Batch research across users

1. POST /pulses with linkType "multi"
2. Send same URL to many users
3. GET /pulses/:id/conversations for all responses

One-off intake for a specific person

1. POST /pulses with linkType "single"
2. Send the URL to that one person
3. Link becomes inactive after they respond

CLI (Alternative to API)

Published on npm as @unformal/cli. Two ways to use it:

# One-off (always latest)
npx @unformal/cli <command>

# Or install globally once, then run as `unformal`
npm i -g @unformal/cli
unformal <command>

Commands:

unformal init --key unf_YOUR_KEY
unformal create --intention "Qualify leads" --mode interview --research
unformal list
unformal get PULSE_ID
unformal update PULSE_ID --fields "score:number:Lead score 1-10"
unformal conversations PULSE_ID    # Summary table of all conversations
unformal conversation CONV_ID      # One conversation: full transcript, structured answers, echo
unformal export PULSE_ID --format json --output all.json   # All raw responses in one file
unformal resonance PULSE_ID        # Aggregate insights (themes, NPS, per-question stats, quotes)
unformal analytics PULSE_ID        # Completion rate, duration, field coverage
unformal usage

Every command supports --json for piping structured output into another agent.

Credits

• 50 free conversations on signup
• Each conversation costs 1 credit (2 with allowResearch)
• Buy more at unformal.ai/studio/settings

Full Documentation

• API Reference: https://unformal.ai/docs/api
• Integration Guide: https://unformal.ai/integrate
• llms.txt: https://unformal.ai/llms.txt