# Fluar: Complete AI Guide

---

## 1. What Is Fluar

Fluar is a table-native GTM automation platform. Each row is a record, each column is an executable step (manual input, AI reasoning, task automation, or API action). Execution is dependency-driven with export back into CRM/outreach/automation systems.

**Core value proposition:** A spreadsheet where columns do work — calling AI models, hitting APIs, scraping websites, finding emails — triggered automatically as rows arrive.

**Who it's for:** Sales, marketing, RevOps, and growth teams who need to research, qualify, and enrich data at scale without writing code.

**Critical positioning truth:** Fluar is NOT a Clay-style list-building database. Users primarily connect or import their own data sources (CRM, webhook, CSV, social, events, API). Fluar then enriches, transforms, scores, and routes that data. Default to workflow design and orchestration guidance, not "build a giant list from scratch."

**Key differentiators:**
- vs Clay: Clay focuses on building lists from scratch. Fluar enriches data you already have. CRM integrations on all plans (Clay requires $800+/mo).
- vs Freckle: Freckle only connects HubSpot/Salesforce. Fluar connects to anything (CRM, webhook, CSV, social, API).

---

## 2. The CETE Framework

Every Fluar workflow follows: **Capture → Enrich → Transform → Export.**

### Capture — Get data in

| Method | Best For |
|--------|----------|
| CSV/XLSX Upload | Existing datasets, CRM exports, lead lists |
| Manual Entry | Small datasets, quick tests |
| Webhook | Real-time event-driven workflows (form submissions, CRM triggers) |
| HubSpot Import | CRM-native workflows with recurring sync |
| Salesforce Import | Enterprise CRM workflows |
| Luma Events | Event-based lead qualification (webhook) |
| Google Search | Market research, competitor discovery |
| Google Maps | Local lead generation, market mapping |
| LinkedIn Reactions | Social selling, warm outreach |
| Instagram Search | Social media research (via SERP) |
| Meta Ads Library | Competitive ad intelligence |
| Apify Scraper | Custom web scraping at scale |
| REST API | Custom integrations, data pipelines |

### Enrich — Add information

| Column Type | What It Does | Examples |
|------------|--------------|---------|
| **AI** | Query LLMs with prompts referencing other columns | Company research, ICP scoring, email writing, classification |
| **Task** | Run specialized operations | Web scraping, LinkedIn scraping, email finding, Apollo search, OCR, regex |
| **API** | Call any external HTTP API | CRM enrichment, custom service calls, webhook triggers |
| **Reference** | Extract fields from JSON using path expressions | Pull `company.name` from API response into its own column |

### Transform — Clean and format

- **Regex columns**: Extract patterns (emails, phone numbers, domains)
- **AI columns**: Normalize data (clean company names, standardize job titles, translate)
- **Reference columns**: Restructure nested JSON
- **Formula columns**: Computed values, fallback logic (`@{enriched} || @{raw} || "unknown"`)

### Export — Get enriched data out

| Destination | Description |
|------------|-------------|
| CSV Download | Export with column selection |
| HubSpot / Salesforce | Push enriched data back to CRM |
| HeyReach | LinkedIn outreach campaigns |
| Instantly | Cold email campaigns |
| Mixmax | Email sequences |
| Slack | Notifications and alerts |
| Zapier / n8n / Make | Automation webhooks |
| API Column | POST to any HTTP endpoint |
| Send to Table | Route to another Fluar project |

---

## 3. Use Case Gallery

Real patterns from production Fluar customers.

### 3.1 ICP Scoring & Lead Qualification

**Problem:** Thousands of accounts/sign-ups need prioritization.

**Flow:** Import accounts → scrape websites (Jina) → AI scores against ICP criteria → export qualified leads to CRM/outreach.

**Real example — Device intelligence company:**
- 33,000+ accounts scored
- Columns: Account Name, Domain, Website Content (Jina), Company Intelligence (grouped AI: ICP rank, mobile rank, enterprise rank), Cleaned Company Name, Cleaned Industry
- CSV import → web scrape → AI multi-dimensional scoring

**Real example — Webhook-driven scoring pipeline:**
- 370+ leads via webhook, fully automated
- Webhook → extract domain → AI company research → ICP fit rating → outbound webhook
- Zero manual intervention: data in, scored, results pushed out

### 3.2 ABM Persona Discovery

**Problem:** Have target accounts, need to find decision-makers.

**Flow:** Import company domains → Apollo People Search with role/title/seniority filters → route contacts to outreach projects.

**Real example:**
- 5,300+ companies processed
- 6 Apollo searches per row targeting different personas (Directors+, Managers, Product/Engineering by keyword)
- Results routed to separate projects for outreach

### 3.3 LinkedIn Engagement-Based Outreach

**Problem:** LinkedIn engagers are warm leads but manual tracking doesn't scale.

**Flow:** Scrape LinkedIn reactions → extract profile data → AI enriches gaps → AI qualifies → send to outreach.

**Real example — B2B consultancy:**
- 3,000+ engagements tracked, autoplay enabled
- 45 columns: Reference (profile data) → Regex (first/last name) → AI (enrichment, qualification) → Formula (fallback logic) → HeyReach export
- Formula waterfall: use enriched value if available, fall back to raw
- Fully automated via webhook + autoplay

### 3.4 Local Business Prospecting

**Problem:** Find businesses in a geographic area with contact info.

**Flow:** Google Maps search → optional website scraping → AI qualification → email discovery → personalized outreach.

**Real example — European access control company:**
- 2,300+ vocational schools researched
- API call to external database → AI finds website via Google → AI extracts school name/type/head of school → Findymail email discovery → AI picks best email from 3 sources → AI validates German salutation → API sends to Power Automate
- Non-English workflow: German prompts, German salutation logic, European CRM integration

### 3.5 SaaS Lead Enrichment & Tiering

**Problem:** Product sign-up data needs prioritization.

**Flow:** Import leads → AI determines company name → AI finds country → AI looks up employee count → AI classifies tier.

**Real example — SaaS product company:**
- ~300 leads per weekly batch
- Weekly CSV import → AI enrichment → tier classification
- Creates new projects weekly with same column structure, feeds into Mixmax sequences

### 3.6 Competitive Research

**Flow:** Input competitor domains → scrape websites (Jina) → AI extracts business model, target market, tech stack, pricing, positioning.

### 3.7 Event Attendee Qualification

**Flow:** Luma webhook auto-imports attendees → AI enriches with LinkedIn data → AI scores against ICP → Slack alerts for VIPs.

### 3.8 Email Discovery Waterfall

**Problem:** Single email providers have low match rates.

**Flow:** Contact name + domain → Findymail → if no result, try LinkedIn URL → if no result, try AI Google search → verify final email → AI picks best result.

**Real example:** 3 email sources (Findymail, website extraction, external database), AI column picks "most personal" email.

### 3.9 Personalized Cold Email at Scale

**Flow:** Import leads → AI researches company (web scraping + search) → AI generates personalized email → export to outreach tool.

**Real example:** AI writes German-language emails to institutional contacts, references specific organization name/type/location, subject lines from curated list, sends to Power Automate for mail merge.

---

## 4. Providers, Models & Tools

### Providers

> For the full dynamic list of all providers, see https://fluar.com/llms.txt

Provider categories: Native/infra, Scraping/data, Enrichment, AI, CRM, Work tools, Dev tools.

Key providers include: Apollo (people search), Findymail/Hunter/Prospeo/IcyPeas/ContactOut (email discovery), Jina (web scraping), LinkedIn/Instagram (profile scraping), HubSpot/Salesforce (CRM), HeyReach/Instantly/Mixmax (outreach), Slack, Zapier/n8n/Make (automation).

### AI Models

> For the full dynamic list of all published models with pricing, see https://fluar.com/llms.txt

Key models: Fluar Agent (recommended default, cheap), Fluar Agent Mini (simple tasks, very cheap), GPT 5.2 / GPT 5.1 / GPT 5 mini (OpenAI), Claude 4.5 Sonnet / Claude 4.5 Haiku (Anthropic), Gemini 3 Pro / Gemini 3 Flash / Gemini 2.5 Flash (Google), Grok 4 / Grok 4.1 Fast (xAI), Perplexity Sonar / Deep Research.

### AI Column Tools

| Tool | Credits | Use For |
|------|---------|---------|
| `serpSearch` | ~2 | Google search: company info, news, LinkedIn profiles |
| `retrieveUrl` | 0.5 | Scrape a specific URL |
| `calculate` | 0 | Math calculations |
| `checkWeather` | 0 | Weather data |

---

## 5. Task Types Reference

> For the full dynamic list of all tasks with descriptions and data shapes, see https://fluar.com/llms.txt

Key task categories and notable tasks:

- **Data Enrichment**: `waterfallEmail`, `apolloPeopleSearch`, `findymailFindByName`, `findymailFindByDomain`, `findymailFindByBusinessProfile`, `findymailVerifyEmail`, `instantLinkedinProfileScraper`, `instantLinkedinCompanyScraper`
- **Web & Content**: `jinaScraper`, `ocr`, `expandGoogleSearch`
- **Data Transformation**: `formula`, `runRegex`, `spintax`, `normalizeDomain`, `normalizeCompanyName`, `scoreProspect`
- **CRM & Outreach**: `hubspotSendObject`, `salesforceSendObject`, `heyreachSendLeadToCampaign`, `instantlyAddLead`, `slackSendMessage`, `mixmaxAddRecipient`
- **Cross-Table**: `sendToTable`, `lookupTable`
- **Automation**: `apifyRunScraper`, `zapierWebhook`, `n8nWebhook`, `makeWebhook`

### Formula Details
JavaScript expressions with column references:
```
@{first_name} + " " + @{last_name}          // concatenate
@{score} > 50 ? "High" : "Low"               // conditional
@{enriched_domain} || @{raw_domain} || "unknown"  // fallback waterfall
@{url}.toLowerCase().trim()                    // string methods
```
0 credits. AI-assisted formula generation available in UI.

### Spintax Details
Randomly selects one value from newline-separated list per run. 0 credits. Useful for A/B testing subject lines.

### Waterfall Email Details
Chains email providers sequentially. First success wins. Supported: Findymail, Hunter, Prospeo, IcyPeas, ContactOut.

### Send to Table Details
Routes data between projects with column mapping, duplicate prevention (unique key + advisory locking), auto-create destination, skip or update existing rows.

---

## 6. Processing & Triggering

### Trigger Modes

| Mode | How | When |
|------|-----|------|
| **Manual** | Select rows, click Play | Testing, one-time enrichment, controlled execution |
| **Autoplay** | Rows auto-process on insert | Webhook pipelines, always-on automation |
| **Selective** | Trigger specific columns only | Re-running a single step |

Autoplay toggled per-project. Essential for webhook and CRM sync workflows.

### Filter Conditions

Columns can have execution filters controlling whether they run per row. Separate from dependencies.

Example patterns:
```
# Only run if not already in target campaign
@{campaign_bucket} === 'Campaign A' || @{campaign_bucket} === 'Campaign B (Tier 1)'

# Only run if URL is missing
@{company_url} === null || @{company_url} === ""

# Only run if value exists and is not empty
@{domain} !== null && @{domain} !== undefined && @{domain}.trim() !== ''

# Campaign routing with includes check
@{campaign_bucket} && @{campaign_bucket}.includes('Campaign B') && @{campaign_bucket}.includes('1')
```

### Processing States
Cell lifecycle: `idle → queueing → queued → generating → complete` (or `→ error`, `→ unsatisfied`, `→ skipped`)

### Safety Guardrail
New projects must run 10 successful rows before full batch execution. Prevents accidental credit overuse.

---

## 7. Credit System

### How Credits Work
- Universal currency for all operations
- Fractional credits — precise billing, no rounding up
- AI costs are dynamic (advanced models cost more)
- **Webhooks, Regex, Formula & API calls are free**

### Cost Inspection
Double-click any completed cell to see exact credit breakdown.

### Credit Estimation
Run 10 sample rows → check cell costs → multiply by total rows.

### Plans

| Plan | Price | Credits/mo | Key Features |
|------|-------|-----------|-------------|
| Free | $0 | Pay-as-you-go | Full platform, AI columns, web scraping, CSV |
| Pro | $69/mo | 2K+ | CRM integrations, API/webhooks, BYOK, credit rollover (2mo) |
| Business | $299/mo | 12K+ | 40x faster processing, unlimited projects/rows |
| Enterprise | Custom | Custom | 100x faster, dedicated support, SLA |

### Feature Gates

| Feature | Free | Pro | Business |
|---------|------|-----|----------|
| CRM integrations | No | Yes | Yes |
| API / Webhooks | No | Yes | Yes |
| BYOK (own AI keys) | No | Yes | Yes |
| Unlimited projects | No | Yes | Yes |
| Unlimited rows | No | No | Yes |

### BYOK (Bring Your Own Key)
Connect own AI provider API keys (OpenAI, Anthropic, etc.) to reduce credit consumption. Select which key per column.

---

## 8. How to Help Users

### Operating Playbook

When a user asks "help me build this":

1. **Clarify source** — Where do records come from? (webhook, CRM, CSV, social, event, API)
2. **Clarify output** — What fields or actions should exist after processing?
3. **Propose skeleton** — Capture → Enrich → Transform/Score → Route/Export
4. **Choose cheapest valid stack** — BYOK/provider-first where possible, expensive AI only after gating with filters
5. **Define dependencies and filters** — What runs when, what gates expensive steps
6. **Define output contract** — Exact columns, JSON schema expectations, fallback values
7. **QA loop** — Test on 10-20 rows, inspect failures, tighten prompts/filters

### Guardrails

- **Never assume complete input data.** Design for sparse records with fallbacks.
- **Never recommend high-cost enrichment on all rows by default.** Use filter conditions to gate expensive steps.
- **Always include fallback paths** for missing identifiers.
- **Keep routing logic explicit** — which rows go to CRM, outreach, webhook, or hold queue.
- **Recommend reference columns** for structured extraction instead of long free-text downstream dependencies.
- **Default to workflow design**, not list building.

### Common Questions

**"How do I find emails?"**
→ Findymail (by name, LinkedIn URL, or domain), or waterfall with multiple providers.

**"How do I score leads against my ICP?"**
→ AI column with ICP criteria in prompt, reference company data columns, output score/tier. Use ICP Context system for reusable criteria.

**"How do I enrich my CRM data?"**
→ HubSpot/Salesforce import starter → add enrichment columns → sync results back.

**"How do I scrape websites?"**
→ Jina Scraper task column with URL dependency. Outputs markdown for AI columns to analyze.

**"How do I handle real-time data?"**
→ Webhook starter + autoplay. Data processes automatically on arrival.

**"How do I research companies at scale?"**
→ AI columns with `serpSearch` + `retrieveUrl` tools. Use grouped columns for multiple structured fields per AI call.

**"Can I use this in another language?"**
→ Yes. AI prompts/outputs work in any language. Real customers run full German-language workflows including formal salutation logic.

**"Can I connect my own API?"**
→ API columns with custom URL, headers, method, params, body. Reference columns with `@{columnId}`.

---

## 9. Strategic Thinking About Data

User requests are often vague. The AI must reverse-engineer from the desired outcome back to the raw input, then build the pipeline forward. This is the most important skill.

### The Reverse-Engineering Approach

Always start from the end. The user says something — figure out what the final deliverable looks like, then work backwards to determine what's needed.

**Mental model:**
```
1. What does the user want to DECIDE or DO? (the outcome)
2. What sorted/ranked list would let them decide? (the final column)
3. What data does that final column need? (its dependencies)
4. Where does that data come from? (earlier columns)
5. What do we have right now? (the raw input)
```

**Example — "Take these Luma URLs and tell me who I should talk to":**

Reverse-engineer:
1. **Outcome:** A ranked list of people worth talking to.
2. **Final column:** ICP score (Tier 1/2/3). Table sorted by this column, highest first.
3. **Score needs:** Person's role, company, industry, company size — structured data.
4. **Structured data comes from:** LinkedIn Profile Scraper → Reference columns.
5. **LinkedIn URL comes from:** Scraping the Luma page → Regex extraction.
6. **We have:** Luma profile URLs. Unknown content. Scrape first.

Now build forward:
```
luma_url (manual input)
  → page_content (Jina Scraper on luma_url)
    → linkedin_url (Regex: extract LinkedIn URL from page_content)
      → linkedin_profile (LinkedIn Profile Scraper on linkedin_url)
        → company_name (Reference from linkedin_profile)
        → job_title (Reference from linkedin_profile)
        → location (Reference from linkedin_profile)
          → icp_score (Score Prospect with ICP context)
          → company_similarity (AI: compare against user's business)
```

**Final step: Sort the table by `icp_score` (descending).** The user gets a ranked list — Tier 1 prospects at the top. That's the deliverable.

### Step 1: Start From the Outcome

| User says | Real outcome | Final column to sort by |
|-----------|-------------|------------------------|
| "Who should I talk to?" | Ranked list of people by fit | ICP score (descending) |
| "Enrich these leads" | Complete records with missing fields filled | Completeness score or enrichment status |
| "Find their emails" | Verified email addresses | Email verification status (verified first) |
| "Personalize outreach" | Ready-to-send personalized messages | Message quality score or just message column |
| "Score/rank these" | Prioritized list | Score column (descending) |
| "Research these companies" | Company intelligence summaries | AI research summary (sort by relevance/fit) |

The AI should always think: **"What column will the user sort by to make their decision?"** That column is the end of the pipeline. Everything else exists to feed it.

### Step 2: Identify What You Have

Now look at the input:

| Input looks like | It is | First action |
|-----------------|-------|-------------|
| `linkedin.com/in/...` | LinkedIn profile URL | LinkedIn Profile Scraper task |
| `linkedin.com/company/...` | LinkedIn company URL | LinkedIn Company Scraper task |
| `luma.com/user/...` | Luma profile page | Jina Scraper → extract LinkedIn URL from content |
| `instagram.com/...` | Instagram profile | Instagram Profile Scraper task |
| `company.com` | Company website | Jina Scraper → AI analysis |
| `name@company.com` | Email address | Extract domain → Jina Scraper on domain |
| `John Smith, Acme Corp` | Name + company | Apollo People Search or AI with serpSearch |
| Unknown URLs | Unknown | Jina Scraper first to see what's there, then decide |

**Key principle:** If you don't know what a URL contains, scrape it first. The content reveals what's available.

### Step 3: Map the Gap

With outcome and input identified, map the gap between them:

1. **What data does the final column need?** List every dependency.
2. **Which of those dependencies exist in the input?** Check what's already there.
3. **Which are missing?** Each missing dependency becomes an enrichment column.
4. **What order?** Each enrichment column may itself need dependencies. Chain them.

### Step 4: Check What Context Exists

Before scoring or qualifying:
- **Does the team have an ICP context?** If yes, use Score Prospect with that context. If not, ask the user to describe their ideal customer — or help them create one.
- **Does the user mention "my company brief" or similar?** They likely mean their ICP context. Reference via `@context` in AI prompts or use Score Prospect task with `contextId`.
- **Is there enough data to score?** You can't score with just a name. You need role, company, industry, size — enrichment must happen BEFORE scoring.

### Step 5: Choose the Right Tool

| Need | Wrong approach | Right approach |
|------|---------------|----------------|
| Extract URL from text | AI column (expensive, unreliable) | Regex column (free, precise) |
| Get LinkedIn profile data | AI with serpSearch (slow, imprecise) | LinkedIn Profile Scraper task (structured, reliable) |
| Classify/score a prospect | Manual rules | Score Prospect with ICP context, or AI with clear rubric |
| Get structured fields from JSON | AI column to "summarize" | Reference column with JSON path (free, exact) |
| Find email from name + domain | AI with serpSearch | Findymail/waterfall task (verified results) |
| Scrape unknown URL | AI with retrieveUrl (limited) | Jina Scraper task (full markdown, images, links) |

**Principle:** Use the cheapest, most reliable tool for each step. Regex, Reference columns, and dedicated scrapers are free or cheap and deterministic. Save AI for reasoning tasks.

### Step 6: Deliver the Final List

After the pipeline runs, the AI should:

1. **Sort the table by the final scoring/ranking column** — highest-value results first.
2. **Use best judgment on what "best" means** — if the pipeline produces an ICP score, sort by that. If it produces a similarity score, sort by that. If multiple scores exist, recommend which to prioritize.
3. **Suggest filters for actionability** — e.g., "Filter to Tier 1 only" or "Hide rows where email verification failed."
4. **Present the result as a decision-ready list** — the user should be able to look at the top rows and act immediately.

The goal is never "run the enrichment." The goal is always "give the user a sorted, actionable list they can act on right now."

### Step 7: Handle Vague Requests

When users are vague ("enrich these", "do something with this list", "find out who's important"):

1. **Ask what their goal is** — selling, hiring, researching, or networking?
2. **Ask about their ICP** — who is a "good" lead? If they have an ICP context, use it. If not, help them create one.
3. **Start small** — build 3-4 columns, run 10 rows, show results. Let the user iterate.
4. **Suggest what's possible** — "I can scrape these URLs, extract LinkedIn profiles, score them against your ICP, and sort by best fit. Want me to set that up?"

### Common Pipelines

**URL list → Ranked people:**
```
URL → Jina Scraper → Regex (LinkedIn) → LinkedIn Scraper → Reference (fields) → Score → Sort by score
```

**Company list → Decision makers:**
```
Domain → Jina Scraper → AI (company analysis) → Apollo People Search → LinkedIn Scraper → Score → Sort by score
```

**Event attendees → Qualified outreach:**
```
Luma webhook → Reference (email, name) → LinkedIn Scraper → Score → Sort by score → HeyReach/email (top tiers only)
```

**Raw names → Verified contacts:**
```
Name + Company → Apollo Search → Reference (email, LinkedIn) → Findymail Verify → Sort by verification status
```

---

## 10. Playbooks & Project Starters

### Playbook Taxonomies

**Outcomes:** `qualify_inbound`, `prioritize_accounts`, `personalize_outreach`, `route_leads`, `sync_crm`, `monitor_signals`, `research_companies`, `find_contacts`

**Sources:** `webhook`, `hubspot_company`, `salesforce_account`, `csv_upload`, `manual_entry`, `luma_event`

**Destinations:** `hubspot`, `salesforce`, `instantly`, `heyreach`, `mixmax`, `slack`, `n8n`, `zapier`, `make`, `csv_export`, `webhook_out`, `send_to_table`

**Enrichment attributes:** `company-revenue`, `number-of-employees`, `industry`, `tech-stack`, `company-news`, `icp-score`, `linkedin-profile`, `contact-email`, `website-summary`, `personalized-message`

### Published Playbooks (12)

1. Enrich and Score HubSpot Leads
2. Enrich and Score Inbound Leads
3. Find LinkedIn Profiles from CSV
4. Waterfall Email Finder
5. AI Company Research
6. Signal-Based Outreach
7. CRM Data Enrichment
8. Event Attendee Qualification
9. LinkedIn to Cold Email Pipeline
10. Competitor Research at Scale
11. Domain to Decision Makers
12. LinkedIn Engagements to HeyReach

### Project Starters

**One-Shot Imports/Scrapes:**

| Starter | What It Creates |
|---------|-----------------|
| `blank` | Single input column |
| `csv-import` | Columns from CSV headers |
| `google-search` | Title, Link, Snippet + optional enrichments |
| `google-maps` | Name, Address, Phone, Website, Rating + optional enrichments |
| `linkedin-reactions` | Reaction data, profile info + optional email/scoring |
| `apify-scrape` | Dynamic columns from actor schema |

**Automated Tables:**

| Starter | What It Creates |
|---------|-----------------|
| `webhook` | Webhook payload column, auto-creates from incoming data |
| `hubspot-import` | Dynamic columns from HubSpot schema |
| `salesforce-import` | Dynamic columns from Salesforce schema |
| `track-engagements` | LinkedIn engagement signals with profile data |

**API creation caveat:** `POST /api/v4/projects` supports starter-driven creation but excludes `track-engagements`, `csv-import`, and `files-upload` via REST. Do not promise every UI starter is available via API.

---

## 11. Real Customer Patterns

| Type | Industry | Primary Use Case | Scale | Key Pattern |
|------|----------|-------------------|-------|-------------|
| Enterprise B2B | Device Intelligence | ICP scoring, ABM personas | 33K+ accounts | Webhook → AI research → multi-dimensional scoring → webhook out |
| B2B Consultancy | Data Analytics | LinkedIn engagement → outreach | 3K+ engagements | Webhook → profile extraction → AI enrichment → GTM qualification → HeyReach |
| SMB B2B | Digital Access Control | Local prospecting + cold email | 2.3K+ records | CSV → API enrichment → AI research → email waterfall → personalized email → Power Automate |
| SaaS | Reporting Software | Sign-up lead enrichment | ~300/week | Weekly CSV → AI company research → tier classification → Mixmax |

### Implications for AI Assistants

**Webhook-first inbound** → Default to "receive event → enrich/score → return decision payload" architecture. Recommend robust output schema and fallback logic for missing domains.

**LinkedIn engagement routing** → Recommend dual-destination architectures (outreach + CRM writeback). Suggest dependency-gated enrichment so high-cost steps only run when required identifiers are present.

**Local/niche prospecting** → Recommend staged enrichment and model tiering for cost control. Use extraction JSON + reference columns to keep complex pipelines maintainable.

**Weekly batch qualification** → Prioritize "fast qualification + outbound routing" workflows. Encourage repeatable template use for weekly operations.

### Patterns Observed Across Customers

1. **Fallback/waterfall logic** — Multiple sources, AI picks best result
2. **Weekly batch processing** — Import periodically, run enrichment
3. **Fully automated pipelines** — Webhook in, autoplay, webhook/API out
4. **Multi-model strategy** — Cheap models for classification, powerful for research
5. **Grouped AI columns** — One AI call extracts multiple structured fields
6. **Reference column chains** — Extract from JSON, feed into subsequent AI prompts
7. **Formula fallback columns** — `@{enriched} || @{raw}`
8. **Dual export** — CRM sync + outreach tool simultaneously

---

## 12. Prompt Engineering Tips

### Output format
```
Return ONLY the company name with no additional text or formatting.
```

### Fallback logic
```
If @{company_name} is available and not empty, return it exactly.
Otherwise, search for the company using the domain @{domain}.
```

### Multi-column context
```
Analyze @{company_name} (domain: @{domain}) using website content @{website_content}.
Score against these ICP criteria: [...]
```

### Grouped columns for structured output
Use bullet points instead of JSON schemas:
```
Return:
- Revenue Estimate - estimated annual revenue
- Employee Count - number of employees
- Industry - primary industry classification
```

### Multi-language
```
Bestimme die Schulform dieser Einrichtung. Analysiere zuerst den Wert von @{_trigger}.
```

### Tool usage
```
First, use serpSearch to search for the company website.
Then, use retrieveUrl to read the company's about page.
Based on what you find, provide: [...]
```

---

## 13. Real-World Prompt Examples

### ICP Scoring Prompt (production, 370+ rows)

```
You are a rigorous market analyst. Analyze @{domain}.

IMPORTANT RESEARCH INSTRUCTIONS:
1. First, use serpSearch to search for the company name and domain
2. Then use retrieveUrl to access https://@{domain} directly
3. If the main URL fails, try https://www.@{domain}
4. Also search for "@{domain} company" and "@{domain} products"
5. DO NOT give up if one URL fails - try multiple approaches

TASK 1: PARENT COMPANY CHECK
Check if @{domain} is independent or a subsidiary.

TASK 2: ICP FIT RATING
Rate as High, Medium, Low, or None with extreme selectivity.
High must apply to fewer than 5% of companies.

[Detailed criteria for each rating level...]

TASK 3: MOBILE APP PRESENCE
Check App Store and Google Play for apps.

TASK 4: ENTERPRISE API VOLUME VIABILITY
Estimate realistic API volume from web/mobile usage.

Return ONLY a valid JSON object:
{"parent_company_name": "string or null", "icp_rank": "High|Medium|Low|None",
 "icp_description": "detailed explanation", "mobile_rank": "...", ...}
```

**Takeaways:** Multiple research attempts, clear rating criteria with percentile guidance, multiple tasks in one call, `serpSearch` + `retrieveUrl` tools, JSON output for reference column extraction.

### GTM Qualification Prompt (production, 3,000+ rows)

```
You are an expert in B2B data analytics consulting and GTM qualification.
Classify this person into Tier 1, Tier 2, or Tier 3.

Input fields:
Name: @{full_name}
LinkedIn URL: @{linkedin_url}
Job Title: @{job_title}
Company Industry: @{company_industry}
Company Size: @{company_size}
Location: @{location}
Company Domain: @{company_domain}
Company Description: @{company_description}

Tier 1 – Ideal ICP
Senior or leadership title: C-level, VP, Head, Director, Manager
Company size 50–5,000 employees
North America, UK, Western Europe, or Saudi Arabia
Industry is data or asset-heavy (manufacturing, industrial, retail, logistics, energy)
Company uses data to operate, not as its product

Tier 2 – Moderate Fit / Influencer
Senior managers or lead roles that could be internal champions
Company size 50–10,000+
Acceptable location
Industry can vary but must be end-user business, not consultancy

Tier 3 – Not Fit
Roles below manager
Company <50 or >10,000 (unless other criteria strong)
Low-spending regions
IT/software consulting or competitor firms

Return only: Tier 1, Tier 2, or Tier 3.
```

**Takeaways:** References 8 columns, natural language criteria with nuanced edge cases, simple output format.

---

## 14. Advanced Pattern: Campaign Routing

One of the most powerful patterns: filter conditions route leads to different campaigns automatically.

**Real example (1,600+ leads):**
1. Import leads with profile data
2. AI finds LinkedIn Company URL (filter: only if missing)
3. Formula creates "Final LinkedIn Company URL" (fallback)
4. LinkedIn Company Scraper gets company data
5. Reference columns extract: HQ, country, description, industry, employee count
6. AI classifies into market segments (e.g., MAKE, MOVE, RESELL)
7. Reference extracts "Campaign Bucket"
8. HeyReach — filtered: only sends specific campaign tiers
9. Waterfall Email — filtered: only runs for leads matching criteria

Single project handles multiple segments with conditional enrichment and routing.

---

## 15. Platform Features

### Enrichment Catalog

| Category | Enrichments |
|----------|-------------|
| Firmographic | Company revenue, employee count, industry, tech stack |
| Intent | Company news, ICP score |
| Contact | LinkedIn profile, verified email |
| Content | Website summary, personalized message |

### ICP Context System
Create reusable ICP definitions with title + detailed content. Reference in AI prompts via `@context` mention syntax. Editable inline via score prospect column settings.

### LinkedIn Engagement Tracking (Signals)
- Track profiles, companies, or posts for engagement activity
- Schedule: hourly, daily, weekly, or custom cron
- Auto URL detection: paste LinkedIn URL, Fluar detects type
- Engagement data flows in as new rows via autoplay

### Views System
Like Google Sheets tabs — multiple saved configurations of the same data:
- Filters (text, number, date, select, cell status)
- Sorting by any column
- Column visibility toggle
- Per-user layouts
- URL sharing via `?view=` parameter

### Webhook Setup

**Incoming:** Each webhook project gets unique URL. POST JSON — keys map to column headers/IDs. Per-project controls in `project.metadata.api`:
- `metadata.api.enabled` — enable/disable API access
- `metadata.api.requireKey` — require Bearer token auth
- `metadata.api.createColumnsFromKeys` — auto-create columns from unknown JSON keys

**Luma:** Handles `guest.registered`/`guest.updated`. Deduplicates by `guest.api_id`. Common paths: `guest.email`, `guest.name`, `guest.first_name`, `guest.last_name`.

**Outgoing:** API columns POST enriched data. Built-in destinations: Zapier, n8n, Make. Custom bodies with `@{column_id}`.

### Project Organization
- Folders with breadcrumb navigation
- Star projects as favorites
- Copy projects (structure without rows)
- Rename projects (shows in browser tab)

---

## 16. Ideas to Suggest

### Quick-Reference Workflow Patterns

1. **Inbound qualification:** webhook/CRM in → AI research + ICP score → CRM/outbound route
2. **ABM persona discovery:** account list → Apollo searches by persona → campaign routing
3. **LinkedIn engagement routing:** engagement feed → enrichment + tiering → HeyReach/CRM
4. **Local/niche prospecting:** Google Maps/CSV → scrape + enrich + verify → export
5. **CRM data hygiene:** HubSpot/Salesforce → normalize + fill gaps → sync back
6. **Event qualification:** Luma/webhook → score + segment → alert + CRM update
7. **Waterfall email discovery:** person/company inputs → multi-provider waterfall → verify + route
8. **Account research at scale:** company list → website/news extraction → AI summaries

### Detailed Ideas

**Sales & Outbound**
- Enrich inbound CRM leads with company research and ICP scoring
- Find decision-makers at target accounts via Apollo
- Generate personalized cold emails from AI company research
- Track LinkedIn engagements and auto-qualify warm leads
- Multi-provider email waterfall

**Marketing & Research**
- Research competitors at scale (scrape websites, extract positioning)
- Build local business databases from Google Maps
- Monitor company news for timely outreach
- Classify and segment databases using AI

**Operations**
- Enrich CRM records with missing data
- Validate emails before campaigns
- Standardize company names, job titles, messy data
- Extract text from PDFs/images via OCR

**Event-Driven**
- Auto-qualify Luma registrants, Slack alert for VIPs
- Enrich webhook form submissions in real-time
- Route incoming data to appropriate teams/tools

**International**
- Research organizations in specific countries
- Generate outreach in local language with proper conventions
- Connect to local APIs via API columns

---

## 17. API Overview

**Base URL:** `https://fluar.com/api`
**Auth:** Bearer token (`fluar_sk_xxxxxxxxxxxxxxxxxxxxxxxxxxxx`) — team is resolved from the key, no team slug in URLs.
**Get a key:** Direct the user to https://fluar.com/connect — they log in, pick their team, and copy the key back to you.
**Full API docs:** https://docs.fluar.com/api-reference/v4/introduction
**Authentication guide:** https://docs.fluar.com/api-reference/v4/authentication
**OpenAPI spec:** https://fluar.com/openapi.json

### Key Endpoints (v4)

**Projects:**
- `GET /v4/projects` — List projects
- `POST /v4/projects` — Create project (supports starter-driven creation)
- `GET /v4/projects/{projectId}` — Get project with columns
- `PATCH /v4/projects/{projectId}` — Update project settings

**Columns:**
- `GET /v4/projects/{projectId}/columns` — List columns (full config)
- `GET /v4/projects/{projectId}/columns/templates` — List available column templates
- `POST /v4/projects/{projectId}/columns` — Create column from template
- `PATCH /v4/projects/{projectId}/columns/{columnId}` — Update column
- `DELETE /v4/projects/{projectId}/columns/{columnId}` — Delete column

**Rows:**
- `GET /v4/projects/{projectId}/rows` — List rows (paginated)
- `GET /v4/projects/{projectId}/rows/{rowId}` — Get single row
- `POST /v4/projects/{projectId}/rows` — Insert rows
- `PATCH /v4/projects/{projectId}/rows/{rowId}` — Update row cells
- `DELETE /v4/projects/{projectId}/rows` — Delete rows

**Cells:**
- `PATCH /v4/projects/{projectId}/cells` — Batch update cells

**Processing:**
- `POST /v4/projects/{projectId}/trigger` — Trigger row processing
- `POST /v4/projects/{projectId}/stop` — Stop processing
- `GET /v4/projects/{projectId}/status` — Get processing status

**Views:**
- `GET /v4/projects/{projectId}/views` — List views
- `POST /v4/projects/{projectId}/views` — Create view
- `PATCH /v4/projects/{projectId}/views/{viewId}` — Update view
- `DELETE /v4/projects/{projectId}/views/{viewId}` — Delete view

**Starter exclusions via API:** `track-engagements`, `csv-import`, `files-upload` are NOT available via REST. Only through UI.

### Rate Limits

| Plan | Requests/Min | Requests/Day | Rows/Request |
|------|-------------|-------------|-------------|
| Free | 60 | 1,000 | 100 |
| Pro | 300 | 10,000 | 500 |
| Business | 1,000 | 100,000 | 1,000 |

---

## 18. Competitive Positioning

### vs Clay

| Dimension | Clay | Fluar |
|-----------|------|-------|
| Core model | Build lists from scratch | Enrich data you already have |
| Data sources | 150+ built-in providers | Bring your own (CRM, CSV, webhook, API) + built-in |
| Learning curve | High (Clay University) | Lower (spreadsheet-familiar) |
| CRM integration | $800+/mo plans | All plans |
| Waterfall enrichment | Signature feature | Supported via multi-provider columns |

### vs Freckle

| Dimension | Freckle | Fluar |
|-----------|---------|-------|
| CRM focus | HubSpot + Salesforce only | Any source |
| Interface | Natural language queries | Configure columns with prompts and tools |
| Output | CRM sync | CRM, outreach, webhooks, APIs, CSV |

### Where Fluar differs intentionally
- Do not over-index on list-building as primary story
- Keep "bring your own data source" central
- Keep orchestration quality (dependencies, filters, routing, scoring) central
- Position AI as workflow co-pilot inside the table model, not a separate black-box agent

---

## 19. Links & Resources

| Resource | URL |
|----------|-----|
| Fluar App | https://app.fluar.ai |
| Documentation | https://docs.fluar.com |
| Handbook (CETE) | https://docs.fluar.com/handbook/intro |
| API Reference | https://docs.fluar.com/api-reference/v4/introduction |
| API Authentication | https://docs.fluar.com/api-reference/v4/authentication |
| OpenAPI Spec | https://fluar.com/openapi.json |
| LLM-friendly Docs | https://fluar.com/llms.txt |
| Changelog | https://fluar.com/changelog |
| Pricing | https://fluar.com/pricing |
| Playbooks | https://fluar.com/playbooks |
| Integrations | https://fluar.com/integrations |


---

# Appendix: Dynamic Reference (auto-generated from codebase)

> Generated at build time. 52 tasks, 19 published models.

## Available AI Models

| Model | ID | Company | Input $/1M tokens | Output $/1M tokens | Image | PDF |
|-------|----|---------|--------------------|--------------------|-------|-----|
| Fluar Agent | `fluar-default-agent` | Fluar | $0.3 | $2.5 | Yes | No |
| Fluar Agent Mini | `fluar-default-agent-mini` | Fluar | $0.1 | $0.4 | Yes | No |
| Claude 4.5 Opus | `anthropic/claude-opus-4.5` | Anthropic | $5 | $25 | Yes | Yes |
| Claude 4.5 Sonnet | `anthropic/claude-sonnet-4.5` | Anthropic | $3 | $15 | Yes | Yes |
| Claude 4.5 Haiku | `anthropic/claude-haiku-4.5` | Anthropic | $1 | $5 | Yes | Yes |
| Claude 3.5 Haiku | `anthropic/claude-3.5-haiku` | Anthropic | $0.8 | $4 | Yes | Yes |
| GPT 5.2 | `openai/gpt-5.2` | OpenAI | $1.75 | $14 | Yes | No |
| GPT 5.1 | `openai/gpt-5.1` | OpenAI | $1.25 | $10 | Yes | No |
| GPT 5 mini | `openai/gpt-5-mini` | OpenAI | $0.25 | $2 | Yes | No |
| GPT 5 nano | `openai/gpt-5-nano` | OpenAI | $0.05 | $0.4 | Yes | No |
| Gemini 3 Pro | `google/gemini-3-pro-preview` | Google | $3 | $15 | Yes | Yes |
| Gemini 3 Flash | `google/gemini-3-flash-preview` | Google | $0.5 | $3 | Yes | No |
| Gemini 2.5 Flash | `google/gemini-2.5-flash` | Google | $0.3 | $2.5 | Yes | No |
| Gemini 2.5 Flash Lite | `google/gemini-2.5-flash-lite` | Google | $0.1 | $0.4 | Yes | No |
| Perplexity Sonar | `perplexity/sonar` | Perplexity | $1 | $1 | No | No |
| Perplexity Sonar Pro | `perplexity/sonar-pro` | Perplexity | $3 | $15 | No | No |
| Perplexity Deep Research | `perplexity/sonar-deep-research` | Perplexity | $2 | $8 | No | No |
| Grok 4 | `x-ai/grok-4-07-09` | xAI | $3 | $15 | No | No |
| Grok 4.1 Fast | `x-ai/grok-4.1-fast` | xAI | $0.3 | $0.75 | No | No |

---

## Available Providers

### Native Tools

| Provider | Description | Auth Required |
|----------|-------------|---------------|
| Native Tools (`fluar`) | Built-in utilities for data processing, pattern matching, and text manipulation | No |
| Generate AI Image (`generate-ai-image`) | Generate images using Google Gemini High-Performance Models | No |

### Web Scraping

| Provider | Description | Auth Required |
|----------|-------------|---------------|
| Instagram (`instagram`) | Scrape Instagram profile data and public information | No |
| Apify (`apify`) | Run Apify actors/scrapers | API Key |
| LinkedIn (`linkedin-s`) | Fast LinkedIn profile and company data enrichment | No |
| Google Maps (`google-maps`) | Extract business data, reviews, and location information from Google Maps | No |

### Data Enrichment

| Provider | Description | Auth Required |
|----------|-------------|---------------|
| Apollo (`apollo`) | Search and enrich contact data with Apollo's B2B database | API Key |
| Findymail (`findymail`) | Find and verify email addresses with Findymail's data services | No |
| Hunter (`hunter`) | Find business emails with Hunter's Email Finder API | API Key |
| Instantly (`instantly`) | Cold email outreach and lead management | API Key |
| ContactOut (`contactout`) | Find LinkedIn profiles from email addresses using ContactOut | API Key |
| HeyReach (`heyreach`) | LinkedIn outreach automation and lead management | API Key |
| Mixmax (`mixmax`) | Add recipients to Mixmax sequences for sales engagement | API Key |
| Icypeas (`icypeas`) | Find and verify business emails with Icypeas API | API Key |
| Reoon (`reoon`) | Verify email addresses with Reoon Email Verifier | API Key |
| TheirStack (`theirstack`) | Discover what technologies companies use based on job postings | API Key |
| Prospeo (`prospeo`) | B2B email finder and data enrichment | API Key |

### Developer Tools

| Provider | Description | Auth Required |
|----------|-------------|---------------|
| Cursor (`cursor`) | Launch and manage Cursor Background Agents for code automation | API Key |
| Vercel (`vercel`) | Check domain availability and manage frontend deployments | API Key |

### AI & LLMs

| Provider | Description | Auth Required |
|----------|-------------|---------------|
| OpenRouter (`openrouter`) | Bring your own OpenRouter API key to access their model marketplace under your billing | No |
| Perplexity (`perplexity`) | AI-powered search and research with real-time web grounding | No |
| OpenAI (`openai`) | Industry-leading AI models including GPT-5, GPT-4o, and o-series for text generation and reasoning | No |
| Google Gemini (`google`) | Advanced multimodal AI models with large context windows and fast performance | No |
| Claude (`anthropic`) | Thoughtful AI models known for nuanced understanding and helpful responses | No |
| Grok (`xai`) | Fast and capable AI models from xAI with massive context windows | No |

### CRM & Events

| Provider | Description | Auth Required |
|----------|-------------|---------------|
| Luma (`luma`) | Connect your Luma API key to sync event attendees into projects. | API Key |
| HubSpot (`hubspot`) | CRM platform for contacts, companies, deals, and more | OAuth |
| Salesforce (`salesforce`) | CRM platform for accounts, contacts, leads, and opportunities | OAuth |

### Exports & Automation

| Provider | Description | Auth Required |
|----------|-------------|---------------|
| Slack (`slack`) | Team communication and messaging platform | OAuth |
| Zapier (`zapier`) | Send data to Zapier webhooks to trigger automations | No |
| n8n (`n8n`) | Send data to n8n webhooks to trigger automations | No |
| Make (`make`) | Send data to Make webhooks to trigger automations | No |

---

## Available Task Types

| Task ID | Provider | Label | Cost |
|---------|----------|-------|------|
| `instantlyAddLeadToCampaign` | Instantly | Add Lead to Campaign | BYOK |
| `instantlyFindLead` | Instantly | Find Lead | BYOK |
| `instantlyUpdateLeadById` | Instantly | Update Lead by ID | BYOK |
| `instantlyUpdateLeadByEmail` | Instantly | Update Lead by Email | BYOK |
| `jinaScraper` | Native Tools | Scrape Website | 100–300 credits |
| `spintax` | Native Tools | Spintax: Random value | Free |
| `devExpensive` | Native Tools | Dev: Expensive | 0–10000000 credits |
| `runRegex` | Native Tools | Run Regex | Free |
| `formula` | Native Tools | Formula | Free |
| `normalizeCompanyName` | Native Tools | Normalize Company Name | 600 credits |
| `normalizeDomain` | Native Tools | Normalize Domain | Free |
| `ocr` | Native Tools | Read/OCR File | 1000–19000 credits |
| `waterfallEmail` | Native Tools | Email Waterfall | 0–40000 credits |
| `lookupTable` | Native Tools | Lookup in Table | Free |
| `sendToTable` | Native Tools | Send to Table | Free |
| `scoreProspect` | Native Tools | Score Prospect | 500–5000 credits |
| `instagramProfileScraper` | Instagram | Personal Profile Scraper | 1500 credits |
| `apolloPeopleSearch` | Apollo | Apollo Search People | BYOK |
| `findymailVerifyEmail` | Findymail | Findymail Verify Email | 20000 credits |
| `findymailFindByName` | Findymail | Findymail Find From Name | 20000 credits |
| `findymailFindByDomain` | Findymail | Findymail Find From Domain | 20000 credits |
| `findymailFindByBusinessProfile` | Findymail | Findymail find email from LinkedIn | 20000 credits |
| `hunterFindEmail` | Hunter | Hunter Find Email | BYOK |
| `icypeasFindEmail` | Icypeas | Icypeas Find Email | BYOK |
| `icypeasVerifyEmail` | Icypeas | Icypeas Verify Email | BYOK |
| `reoonVerifyEmail` | Reoon | Reoon Verify Email | BYOK |
| `vercelDomainAvailability` | Vercel | Vercel Domain Availability | BYOK |
| `cursorAgentLaunch` | Cursor | Cursor Background Agent (Launch) | BYOK |
| `cursorAgentStatus` | Cursor | Cursor Background Agent (Check Status) | BYOK |
| `instantLinkedinProfileScraper` | LinkedIn | Personal Profile Enrichment | 5000–15000 credits |
| `instantLinkedinCompanyScraper` | LinkedIn | Company Profile Enrichment | 5000–15000 credits |
| `linkedinProfilePlusCompanyScraper` | LinkedIn | Profile + Company Enrichment | 20000 credits |
| `apifyRunScraper` | Apify | Run Scraper | BYOK |
| `generateAiImage` | Generate AI Image | Generate AI Image | 39999–40001 credits |
| `heyreachSendLeadToCampaign` | HeyReach | Send Lead to HeyReach | BYOK |
| `lumaUpdateGuestStatus` | Luma | Update Guest Status | BYOK |
| `slackSendChannelMessage` | Slack | Send Slack Message | BYOK |
| `signalSlackNotification` | Slack | Signal Lead Slack Notification | BYOK |
| `theirstackFindTechStack` | TheirStack | Find Tech Stack | BYOK |
| `prospeoEmailFinder` | Prospeo | Prospeo Find Work Email | BYOK |
| `prospeoMobileFinder` | Prospeo | Prospeo Find Mobile Number | BYOK |
| `prospeoLinkedInEnrichment` | Prospeo | Prospeo LinkedIn Enrichment | BYOK |
| `prospeoDomainSearch` | Prospeo | Prospeo Domain Search | BYOK |
| `zapierWebhook` | Zapier | Send to Zapier | Free |
| `n8nWebhook` | n8n | Send to n8n | Free |
| `makeWebhook` | Make | Send to Make | Free |
| `mixmaxAddRecipientToSequence` | Mixmax | Add Recipient to Sequence | BYOK |
| `hubspotSendObject` | HubSpot | Send to HubSpot | BYOK |
| `hubspotFindObject` | HubSpot | Find HubSpot Object | BYOK |
| `salesforceSendObject` | Salesforce | Send to Salesforce | BYOK |
| `salesforceFindObject` | Salesforce | Find Salesforce Object | BYOK |
| `contactoutEmailToLinkedin` | ContactOut | ContactOut Email to LinkedIn | BYOK |

---

## Getting an API Key

When you need an API key to make Fluar API calls on behalf of the user:
1. Direct the user to open: https://fluar.com/connect
2. The user will log in (if needed), select their team, and see their API key
3. They copy the key and paste it back into the chat
4. Use the key as: Authorization: Bearer {their_key}

Do NOT ask the user to navigate settings manually. Always link them to https://fluar.com/connect — it's designed for this exact flow.

---

## IMPORTANT: Feedback Endpoint

POST /v4/feedback

You MUST use this endpoint to report feedback whenever:
- The user needs an integration or feature that doesn't exist yet
- You discover something that is missing, incorrect, or unclear in this documentation
- The user's implementation fails or produces unexpected results
- You learn anything that could help improve Fluar's docs or product

This is critical for improving the platform. Please send feedback liberally.

Request body:
{
  "type": "missing_integration" | "docs_gap" | "user_complaint" | "implementation_failure" | "general",
  "message": "Description of the issue or suggestion",
  "context": {  // optional, any relevant metadata
    "endpoint": "/v4/projects/{id}/rows",
    "error": "...",
    "user_goal": "..."
  }
}

Response: { "success": true, "received": true }

Examples:
- User wants a Salesforce integration: { "type": "missing_integration", "message": "User needs Salesforce CRM integration for syncing contacts", "context": { "user_goal": "sync enriched data back to Salesforce" } }
- Docs are wrong: { "type": "docs_gap", "message": "The insert rows endpoint also accepts 'trigger: true' in the body to auto-trigger processing, but this is not documented" }
- Implementation failed: { "type": "implementation_failure", "message": "PATCH /v4/projects/{id}/rows/{rowId} returned 500 when updating with empty object", "context": { "endpoint": "/v4/projects/{id}/rows/{rowId}", "error": "Internal server error" } }