Skip to content
flow8

AI-Powered Workflows

AI, Embedded Natively

Most platforms treat AI as an afterthought—a Zapier action you call if you need to, or a separate tool you manage outside your workflow. flow8 is different. AI is built in as a first-class citizen, just like “send an email” or “fetch a database record.”

You use AI the same way you use any other flowlet: drag it onto your canvas, configure it, and it works alongside your other steps. No API keys to manage, no data pipelines to set up, no vendor-specific syntax to learn.

Why This Matters

When AI is native, you can:

  • Make smarter decisions in workflows. Classification, summarization, extraction, routing—all powered by AI.
  • Reduce manual decision-making. Let AI do the heavy lifting (read a document, understand it, extract key data). Humans approve the result.
  • Iterate fast. Change your AI prompt, rerun the flow, see new results. No engineers needed.
  • Switch AI providers without rebuilding. Using OpenAI? Switch to Claude in 30 seconds. No flow restructuring, no API rewrites.

Multi-Provider AI: No Vendor Lock-In

flow8 supports multiple AI providers, all at the same quality level. You choose based on cost, performance, or preference:

  • OpenAI: GPT-4, GPT-4 Turbo, GPT-3.5. Good for text generation, complex reasoning.
  • Anthropic Claude: Claude 3 Opus, Sonnet, Haiku. Excellent for long-form analysis, nuanced classification, code understanding.
  • Mistral: Smaller, faster models. Good for cost-sensitive, latency-critical workloads.
  • Ollama: Self-hosted open-source models (Llama, Mistral, etc.). For organizations that need to keep data on-premise.

Switch providers via configuration. No flow rebuilding, no testing required (in most cases).

Configuration, Not Code

Set your default AI provider in flow8’s settings:

AI Provider: Anthropic Claude
Model: Claude 3 Sonnet
Temperature: 0.7
Max Tokens: 2000

Now every AI flowlet in your flows uses Claude. Later, change it to GPT-4 and all flows adapt.

AI Flowlet Types

Chat Completion

Ask an AI a question, get an answer. Use cases:

  • Summarization: “Summarize this long email in 2 sentences.”
  • Classification: “Read this customer support email. Is it a billing issue, technical issue, or feature request?”
  • Extraction: “Pull the vendor name, invoice number, and total amount from this invoice text.”
  • Generation: “Draft a professional response to this customer complaint based on our service policy.”
  • Reasoning: “A customer is requesting a refund for a purchase 6 months old. Is this within policy? Explain your reasoning.”

Input is text (or structured data converted to text). Output is the AI’s response.

Example in a flow:

  1. Email arrives.
  2. Chat Completion flowlet: “Classify this email: Is it urgent, high-priority, or routine?”
  3. Router: If urgent, branch to urgent handler. Otherwise, standard handler.
  4. Rest of flow executes based on classification.

Data Extraction

Structured data extraction from unstructured text. AI reads messy input and outputs clean, structured data.

Input: Messy invoice PDF text. Output: Structured JSON with fields like vendor, invoice_number, date, items (array), total_amount.

Use cases:

  • Extract fields from PDFs (contracts, invoices, forms).
  • Parse email text into structured data (sender, subject, urgency, category).
  • Pull attributes from product descriptions (SKU, color, size, price).

Why this is better than regex: AI understands context. If an invoice says “Ship to: John Smith, 123 Main St,” AI knows that’s the ship-to address, not a contact address. Regex would just match patterns and break when format varies slightly.

Embeddings

Convert text into numerical vectors that capture meaning. Used for:

  • Semantic search: “Find emails about contract disputes” returns emails discussing negotiation terms, violations, disagreements—not just literal matches for “dispute.”
  • Clustering: “Group these 100 customer emails by topic.” AI produces 5–7 clusters without you specifying what topics to look for.
  • Similarity matching: “Is this customer support request a duplicate of any existing tickets?” Embeddings compare semantic similarity, not just keyword matches.

Use case in flow8:

  1. Customer support email arrives.
  2. Embeddings flowlet: Convert the email to a vector.
  3. Search existing tickets (also vectorized) for semantic similarity.
  4. Router: If high-similarity ticket exists, merge with that ticket; otherwise, create new ticket.

Agent Loop (MCP)

Advanced: Use an AI agent to think through problems, call tools, and reach conclusions.

flow8 supports the Model Context Protocol (MCP), allowing an AI agent to:

  1. Think about the problem.
  2. Decide what information it needs (call a tool: fetch from database, API call, search).
  3. Analyze the result, decide next step.
  4. Repeat until it reaches a conclusion.

Example: Fraud detection

  1. A transaction arrives.
  2. Agent Loop flowlet: “Analyze this transaction for fraud risk. Use these tools: check_customer_history, check_merchant_risk, verify_address.”
  3. Agent thinks: “I need to understand the customer’s history.”
  4. Agent calls check_customer_history tool (FlowAPI call within flow8).
  5. Agent gets result, thinks: “This customer is new. I should check the merchant.”
  6. Agent calls check_merchant_risk.
  7. Agent gets result, synthesizes: “New customer, reputable merchant, low fraud risk.”
  8. flow8 receives the final verdict and proceeds accordingly.

Advantage: Agent reasoning is transparent (you see the thinking and tool calls in logs). It’s also correct more often than a single AI call, because the agent can gather more information before deciding.

Use Cases: AI-Augmented Workflows

Document Intelligence

Scenario: Law firm receives contracts from clients. Currently, a paralegal reads each one (10–20 minutes), extracts key terms, and flags risk items.

With flow8 AI:

  1. Contract arrives.
  2. AI extracts structured data: parties, effective date, expiration date, payment terms, liability caps, termination clauses.
  3. AI analyzes for red flags: “This contract lacks an indemnification clause” or “Payment terms are 60 days (outside our standard 30-day window).”
  4. AI compares against a template of “standard terms” and surfaces deviations.
  5. Flow sends a structured summary to the reviewing attorney (1 minute review instead of 20).

Outcome: Paralegal focuses on negotiation and advising clients. Routine document review is automated.

Automated Summarization

Scenario: Your executive team needs weekly briefings combining data from Salesforce, QuickBooks, and HubSpot. Currently, an analyst writes these summaries (3 hours/week).

With flow8 AI:

  1. Scheduled flow runs Friday at 4 PM.
  2. Pulls deal data, financial data, customer health scores.
  3. AI is given the data and instructions: “Write a 1-paragraph executive summary highlighting wins, risks, and action items.”
  4. AI produces: “This week we won 3 deals worth $120K (above plan). Cash position is strong at $2.3M. But churn is trending up (3 accounts at risk). Recommend: sales team focus on renewal outreach.”
  5. Flow emails the summary and posts to Slack.

Outcome: Always-on reporting, consistent tone, ready in seconds. Analyst now does strategic analysis instead of data wrangling.

Smart Routing

Scenario: Customer support team receives 50+ emails daily. Currently, support coordinator reads each one and assigns to the right queue (billing, technical, product, other). Some emails are ambiguous and get misrouted.

With flow8 AI:

  1. Email arrives.
  2. AI reads it and returns: category (billing, technical, product, other), urgency (critical, high, normal, low), and confidence score (95%).
  3. Router branches based on category.
  4. If confidence < 80%, routes to a human for review first, then the queue.
  5. If confidence >= 80%, routes directly to the appropriate team.

Outcome: Correct routing 95%+ of the time. Misrouted emails are rare. Support team sees only emails in their domain.

Classification at Scale

Scenario: HR team receives 200+ job applications monthly. A recruiting coordinator categorizes each as “strong fit,” “good fit,” or “weak fit” based on requirements. This is tedious and subjective.

With flow8 AI:

  1. Application arrives (resume + cover letter).
  2. AI is given the job description and scoring rubric.
  3. AI reads the application and scores: “This candidate matches 8/10 requirements. Strong technical fit. Some communication gaps in cover letter. Overall: Good Fit.”
  4. AI also suggests: “Consider for future senior roles; strong infrastructure background.”
  5. Router: Strong Fit → send to hiring manager for interview. Good Fit → add to talent pool. Weak Fit → polite rejection email (auto-composed by AI).

Outcome: Recruiting coordinator focuses on relationship-building and interview scheduling. Triage is consistent and fair.

Data Enrichment

Scenario: Sales team has a list of 500 prospect companies. You want to understand industry, size, and growth stage for each. Currently, this requires manual research or a data broker.

With flow8 AI:

  1. Flow reads the company list.
  2. For each company, Iterator flowlet runs in parallel:
    • AI is given the company name.
    • AI researches publicly available info (website, news, LinkedIn, job postings).
    • AI synthesizes: “Industry: SaaS. Size: 50–100 employees. Growth: Series B, ~$5M ARR. Focus: workflow automation.”
  3. Results stored in a CSV.

Outcome: 500 companies enriched in 1 hour. Sales can now segment and prioritize better.

Content Generation

Scenario: Marketing team wants personalized outreach emails for 100+ prospects. Personalizing each email manually takes hours.

With flow8 AI:

  1. Flow reads prospect list (names, companies, titles, LinkedIn profiles).
  2. For each prospect, Iterator flowlet:
    • AI is given the prospect’s details and your company’s value proposition.
    • AI generates a personalized email: “Hi [Name], I noticed [Company] is expanding their [Department]. We help teams like yours [benefit]. Thought this might be relevant.”
  3. Results output as a CSV with email drafts.
  4. Marketing team reviews and sends.

Outcome: 100 personalized drafts in 1 hour. Marketing team adds final polish and sends; response rates are higher due to personalization.

Conversational Workflows

Scenario: Customer service wants a chatbot that can handle common questions without routing to a human.

With flow8 AI:

  1. Customer sends a message (chat, email, or Slack).
  2. Chat Completion flowlet: “Answer this customer question about [topic] professionally. Use this context: [FAQ, account info, recent interactions].”
  3. AI generates a response.
  4. Router: If AI confidence is high, send response and mark resolved. If low, send to human agent with AI’s draft as a starting point.

Outcome: 70% of questions answered instantly by AI. Other 30% escalate with context for agents, so humans are more productive.

Competitive Advantage: AI + Automation

AI alone is interesting (better insights, faster analysis). Automation alone is powerful (fewer manual steps). But AI + Automation together is transformative.

Why:

  • Amplified decisions: AI makes judgment calls (classify, extract, summarize) at scale. Automation executes the decisions instantly.
  • Reduced human burden: Humans focus on exception-handling and strategy, not routine tasks. AI handles the routine at superhuman speed.
  • Continuous improvement: As you refine AI prompts based on real results, the automation gets better over time.
  • Competitive velocity: Competitors without AI-powered automation are slow. You move fast.

Example: Legal firm processes client intake.

  • Without AI: Paralegal reads intake, manually creates matter in Clio, uploads docs (15 min per intake).
  • With AI + automation: AI classifies case type, extracts data, flow8 creates matter, uploads docs, queues for review (1 min per intake, 90% less manual work).
  • Outcome: Same paralegal capacity handles 10x more intakes. The firm can grow revenue without hiring.

Transparency and Safety

Using AI in workflows raises questions: “Can the AI make mistakes?” “How do I know what the AI decided?” “What if the AI is biased?”

flow8 gives you visibility:

Execution Logs

Every AI step logs:

  • Input (what was given to the AI).
  • Output (what the AI returned).
  • Reasoning (if Agent Loop, the step-by-step thinking).
  • Confidence (if extraction or classification, how confident was the AI?).
  • Timestamp and duration.

You can audit the entire decision trail.

Approval Gates

For high-stakes decisions, use a NeedsApproval flowlet after an AI step:

  1. AI classifies a vendor contract as “high-risk.”
  2. Flow pauses and sends the AI’s analysis to the legal team.
  3. Legal reviews the AI’s reasoning and either approves or overrides.
  4. Flow continues.

Humans stay in charge of high-risk decisions.

Cost Control

AI APIs can get expensive if you’re not careful (token usage, per-request fees). flow8 lets you:

  • Set usage quotas (“max 100K tokens per day”).
  • Monitor costs in real time.
  • Switch to cheaper models for non-critical tasks.

Getting Started with AI

You don’t need AI expertise. flow8’s AI flowlets are configured via UI:

  1. Drag a Chat Completion flowlet onto your canvas.
  2. Enter your prompt (natural English, no special syntax): “Classify this email as urgent, high, or normal.”
  3. Select AI provider (default or override).
  4. Wire the email data as input.
  5. Save and run.

If you want to be more sophisticated, use prompt engineering best practices (be specific, provide examples, set the AI’s role). But flow8 makes it approachable even without AI expertise.

Summary: Intelligence at Scale

flow8’s native AI capabilities let you:

  • Make better decisions faster. Classification, extraction, reasoning built into workflows.
  • Automate judgment calls. Not just repetitive tasks—decisions that require understanding context.
  • Stay vendor-agnostic. Multiple AI providers, same interface, easy switching.
  • Maintain transparency. Every AI step is logged and auditable.
  • Amplify your team. AI handles routine analysis; humans focus on strategy and edge cases.

This is workflow automation for the AI age—not just moving data around, but understanding it, reasoning about it, and acting on it.