AIDiveForge AIDiveForge
AIDiveForge — Workflow Pack Implementation Guide

Product Documentation & Knowledge Base

Auto-generate comprehensive product documentation, API guides, and internal wikis from code and specs—reducing documentation debt and keeping content in sync with product changes.

Difficulty: Advanced Tools: 5 Time Saved: 15-25 hours/month Updated: April 10, 2026
Documentation Large Language Models Coding Assistants Writing Tools Audio & Voice
Tools Required
#ToolRoleWebsite
1 o1 Complex reasoning & technical analysis https://openai.com
2 Cody (Sourcegraph) Code understanding & explanation https://sourcegraph.com/cody
3 Claude Comprehensive content writing https://claude.ai
4 Notion AI Wiki organization & updates https://notion.so/product/ai
5 Play.ht Audio documentation creation https://play.ht
In This Guide

# Product Documentation & Knowledge Base AI Workflow Pack

1Overview

This workflow automatically generates comprehensive product documentation, API guides, and internal wikis from your codebase and product specifications. By combining code analysis, technical writing, and knowledge base organization, you reduce documentation debt and keep all content synchronized with product changes. Teams using this pack typically save 15–25 hours per month on manual documentation creation and updates.

---

2What's in This Pack

1. o1

What it does: o1 is a large language model (LLM)—a type of artificial intelligence trained to understand and generate human language—built by OpenAI. It specializes in complex reasoning tasks like mathematical problem-solving, coding, and technical analysis. Unlike general-purpose AI assistants, o1 can break down intricate problems step-by-step before providing answers, making it exceptionally strong at analyzing complex codebases and technical specifications.

Role in this workflow: o1 performs deep technical analysis of code architecture, identifies undocumented dependencies, and generates structurally sound API documentation from complex source code.

Documentation: o1 Documentation

Note:

---

2. Cody (Sourcegraph)

What it does: Cody is an AI coding assistant that understands your entire codebase in context. Rather than analyzing isolated code snippets, it can see the relationships between files, trace function calls, and understand architectural patterns across your project. This makes it exceptionally useful for answering questions like "where is this API endpoint defined?" and "what's calling this function?"

Role in this workflow: Cody extracts code context, traces dependencies, and explains code functionality to feed into documentation generators.

Documentation: Cody Documentation

Note:

---

3. Claude

What it does: Claude is a large language model (LLM) created by Anthropic, designed for writing, analysis, and creative tasks. It excels at understanding context, maintaining consistent tone across long documents, and producing human-readable content. Claude can handle very long inputs (up to 200,000 tokens, roughly 150,000 words), making it ideal for processing entire codebases and generating cohesive documentation.

Role in this workflow: Claude synthesizes code analysis and technical specs into polished, comprehensive documentation with consistent formatting and tone.

Documentation: Claude Documentation

Note:

---

4. Notion AI

What it does: Notion is a workspace app that combines notes, databases, wikis, and project management in a single platform. Notion AI is an optional add-on feature that lets you generate, edit, and organize content directly within Notion pages. You can use it to auto-format documentation, create summaries, and reorganize wiki structures without leaving the Notion interface.

Role in this workflow: Notion AI structures and organizes generated documentation into a searchable, maintainable internal wiki format.

Documentation: Notion AI Documentation

Note:

---

5. Play.ht

What it does: Play.ht is a text-to-speech (TTS) platform that converts written documentation into natural-sounding audio. Unlike robotic TTS engines, Play.ht uses neural voices trained on real human speech, producing audio that sounds nearly indistinguishable from a human narrator. This is useful for creating accessible audio guides, tutorial videos, and learning content alongside written docs.

Role in this workflow: Play.ht converts finalized documentation into audio format for accessibility and multi-modal learning options.

Documentation: Play.ht Documentation

Note:

---

3Prerequisites

---

4Setup & Integration Guide

6. Setting Up o1

  1. Navigate to https://platform.openai.com/signup and create an OpenAI account or log in to an existing account.
  2. Go to Billing > Overview and add a payment method (credit card required).
  3. Navigate to API keys in the left sidebar and click Create new secret key. Copy the key and store it securely in a password manager or secure notes app.
  4. Verify API access by testing in the OpenAI Playground (https://platform.openai.com/playground). Select o1 from the model dropdown.
  5. Set usage limits to control costs: Navigate to Billing > Usage limits and set a hard limit (e.g., $50/month) to prevent unexpected charges.
Integration — other tools in this pack: If you plan to integrate o1 with Zapier or Make, you will use the API key in the Authentication section of an HTTP module. In Make, add an HTTP > Make a request module. Set URL to https://api.openai.com/v1/chat/completions, set Method to POST, and add a header: Authorization = Bearer YOUR_OPENAI_API_KEY (replace YOUR_OPENAI_API_KEY with your actual key from step 3).

---

7. Setting Up Cody (Sourcegraph)

  1. Navigate to https://sourcegraph.com/cody/invite and sign up for a free Cody account.
  2. After signup, Sourcegraph will prompt you to connect a code repository. Click Connect repository and select your Git platform (GitHub, GitLab, or Bitbucket).
  3. Grant Sourcegraph access to your repositories by approving OAuth (an authentication protocol that securely connects services without sharing passwords). Follow the on-screen prompts from your chosen Git platform.
  4. Wait for Sourcegraph to index your repository (this takes 1–10 minutes depending on repository size). You'll see a progress indicator at the top of the page.
  5. Once indexing completes, open the Cody chat interface by navigating to https://sourcegraph.com/cody or using the Cody extension in VS Code.
  6. Test Cody by asking a simple question about your codebase (e.g., "What does the getUserProfile function do?").
Integration — other tools in this pack: Cody does not have a direct API for automation. Instead, manually extract Cody's analysis by copying responses and pasting them into Claude for documentation synthesis. Alternatively, use VS Code's Cody extension to keep code context and documentation side-by-side while writing.

---

8. Setting Up Claude

  1. Navigate to https://claude.ai and click Sign up to create a Claude account using email or Google/Apple login.
  2. On first login, you will land on the free tier with limited messaging. To unlock full access, click your profile icon in the bottom left and select Upgrade to Claude Pro.
  3. Complete the payment flow and confirm your $20/month subscription.
  4. If using Claude's web interface for this workflow, no further setup is needed. Skip steps 5–7 and proceed to "Connecting to other tools."
  5. (Optional) To integrate Claude via API, navigate to https://console.anthropic.com/account/keys and click Create Key. Copy the API key.
  6. (Optional) Store your Claude API key securely in a password manager or environment variable.
Integration — other tools in this pack: If using Claude via the web interface, you will copy o1 and Cody outputs (as text) and paste them into Claude for documentation synthesis. If using Claude's API (for advanced integration), add an HTTP module in Make: Set URL to https://api.anthropic.com/v1/messages, set Method to POST, and add a header: Authorization = Bearer YOUR_CLAUDE_API_KEY. Include a JSON payload with keys model (set to claude-3-5-sonnet-20241022), max_tokens (set to 4096), and messages (set to an array containing the messages to send).

---

9. Setting Up Notion AI

  1. Navigate to https://notion.so/signup and create a free Notion account.
  2. Create a new workspace or open an existing one.
  3. In the bottom left of your Notion workspace, click your workspace name and select Settings & members.
  4. Navigate to the Plans tab and view your current plan (likely Free).
  5. Click Upgrade and select a plan. The free tier or Pro plan ($10/month) is sufficient; the Notion AI add-on ($4/month) is available on all plans.
  6. Complete the payment flow for Notion AI (+$4/month) and confirm the purchase.
  7. Once activated, create a new page and type "/" to open the command menu. Search for "Ask AI" or "AI" to see available Notion AI commands (e.g., "Write," "Edit," "Summarize").
Integration — other tools in this pack: Notion integrates with Claude via Zapier or Make. Create a Zapier Zap that triggers when a new documentation page is created in Notion, sends the content to Claude's API for refinement, and updates the page with the refined version. Alternatively, copy documentation from Claude and paste it directly into a Notion page, then use Notion AI's Summarize or Edit features to refine further.

---

10. Setting Up Play.ht

  1. Navigate to https://play.ht/signup and create an account using email or Google login.
  2. Complete the signup flow and verify your email address.
  3. Log in to your Play.ht dashboard at https://app.play.ht.
  4. Navigate to Account > Billing and select a paid plan starting at $10/month. Choose a plan that matches your expected monthly documentation volume (e.g., 50,000 words/month for the Starter plan).
  5. Complete payment and confirm your subscription.
  6. Navigate to Account > API Keys and click Generate API Key. Copy the key and store it securely.
  7. Test the API by navigating to the Text to Speech editor, pasting sample documentation text, selecting a voice (preview options first), and clicking Generate. Download the resulting MP3 file.
Integration — other tools in this pack: Use Make or Zapier to automate audio generation. In Make, add an HTTP > Make a request module: Set URL to https://api.play.ht/api/v1/convert, set Method to POST, add a header Authorization = Bearer YOUR_PLAY_HT_API_KEY, and include a JSON payload with keys text (your documentation content) and voice_id (a specific voice identifier from Play.ht's voice list). The response will include a download_url for the generated audio file.

---

5Step-by-Step Workflow

Step 1: Extract Code Context with Cody

Tool: Cody

Action: Open Cody in VS Code or at https://sourcegraph.com/cody. Ask Cody to explain key functions, classes, and API endpoints in your codebase. Prepare a list of questions beforehand (e.g., "What parameters does the /users/{id} endpoint accept?", "What does the PaymentProcessor class do?", "What are the dependencies between the auth and user modules?").

Data Handoff: Copy Cody's responses and paste them into a text file (e.g., cody-analysis.txt) or directly into Claude for the next step.
Pro Tip: Ask Cody to generate code comments and docstring examples as a starting point; this reduces the amount of content Claude needs to write from scratch.

---

Step 2: Analyze Complex Technical Requirements with o1

Tool: o1

Action: Open the OpenAI Playground at https://platform.openai.com/playground and select o1 from the model dropdown. Paste your product specifications, API contract files, or architectural diagrams. Ask o1 to identify documentation gaps, suggest API documentation structure, and flag potential edge cases (e.g., error handling scenarios, rate-limiting behavior).

Data Handoff: Copy o1's analysis output into a new document or send it directly to Claude via copy-paste.
Pro Tip: Use o1 for one deep analysis per product update rather than running it repeatedly; this keeps API costs low while maximizing value per request.

---

Step 3: Synthesize and Write Documentation with Claude

Tool: Claude

Action: Open https://claude.ai and start a new conversation. Paste the Cody code analysis and o1 technical analysis together. Provide a template or style guide (e.g., "Write in a technical but beginner-friendly tone, use code examples, organize by endpoint, include error codes"). Ask Claude to generate complete API documentation, user guides, or internal wiki content.

Data Handoff: Copy the final Claude output (typically formatted as Markdown or plain text) into a text editor or directly into Notion in the next step.
Pro Tip: Ask Claude to generate documentation in Markdown format with frontmatter (metadata at the top), which Notion and other tools can parse automatically.

---

Step 4: Structure and Organize into Wiki with Notion AI

Tool: Notion AI

Action: Create a new Notion page for your documentation. Paste the Claude-generated content into the page. Use Notion's page structure features to organize: create Databases for API endpoints, Toggle lists for code examples, and Synced blocks to avoid content duplication across pages. Use Notion AI's Summarize feature to generate table-of-contents entries for large documents.

Data Handoff: Share the Notion wiki link with your team, or export as PDF from Notion (⋯ > Export > PDF) for external sharing.
Pro Tip: Create a Notion template for new API documentation pages so all future docs follow the same structure without manual setup.

---

Step 5: Generate Audio Versions with Play.ht

Tool: Play.ht

Action: Copy your finalized Notion documentation (select all content, Ctrl+A, and copy). Open https://app.play.ht/text-to-speech. Paste the documentation into the text editor. Select a natural-sounding voice from the preview list (test 2–3 voices first). Click Generate and wait for processing to complete (typically 1–5 minutes for long documents).

Data Handoff: Download the resulting MP3 file. Upload it to your documentation portal, learning management system (LMS), or video hosting platform (e.g., Loom, YouTube) with the Notion link in the description.
Pro Tip: Generate audio versions only for final, stable documentation; re-generate immediately after major updates to keep audio in sync.

---

Step 6: Automate Future Updates with Zapier or Make

Tool: Zapier or Make

Action: (Optional for advanced workflows) In Zapier, create a new Zap with a trigger like "New commit to GitHub repository." Add an action to send the commit diff (code changes) to Claude's API, then automatically create a new Notion page with Claude's suggested documentation updates. Configure a second action to notify your team via Slack when new documentation is generated.

Data Handoff: The Zap automatically handles the data flow between GitHub, Claude, Notion, and Slack without manual intervention.
Pro Tip: Test a Zap with a single trigger before going live to ensure the data format matches at each step (e.g., GitHub sends plain text diffs, Claude expects JSON payloads, Notion API expects specific page structures).

---

6Integration Map

``` Code Repository (GitHub/GitLab/Bitbucket) ↓ [Cody] → Extracts code context & explanations (.txt) ↓ [o1] → Analyzes specs & identifies gaps (.txt) ↓ [Claude] → Synthesizes into polished documentation (.md) ↓ [Notion AI] → Structures into searchable wiki (Notion page) ↓ (two paths from Notion) ├─→ [Manual PDF export] → Share with non-technical stakeholders └─→ [Play.ht] → Converts to audio (.mp3) → Upload to LMS or YouTube ↓ [Zapier/Make] → Optional automation: retrigger on code commits ```

File Format Handoff:

Automation Opportunities:

Manual Handoff Points (and why):

---

7Troubleshooting

Problem

o1 API returns a 401 Unauthorized error. Solution: Verify your API key is correct. Navigate to https://platform.openai.com/api-keys, copy the full key again (ensure no extra spaces), and paste it into your Zapier/Make module or request header. Confirm your OpenAI account has an active payment method in Billing > Overview. If the issue persists, regenerate a new API key in the dashboard.

---

Problem

Cody says "Repository not indexed" or returns incomplete code context. Solution: Wait for Sourcegraph to finish indexing your repository. Navigate to https://sourcegraph.com/your-repo-name (replace with your actual repo) and check the progress bar at the top. Indexing can take 5–20 minutes for large repos. If stuck, disconnect and reconnect your repository in Cody Settings > Connected Repositories > Disconnect > Reconnect.

---

Problem

Claude generates documentation but the tone is too technical or too casual for your audience. Solution: In your Claude prompt, add explicit tone instructions: "Write this API documentation for junior developers. Use plain language, avoid jargon, and include a 'Getting Started' section with copy-paste code examples." If the first attempt doesn't match, ask Claude to "rewrite in a more [technical/accessible/concise] tone" and it will regenerate based on your feedback.

---

Problem

Notion AI features are not appearing in my workspace (no "Ask AI" button). Solution: Confirm you have activated the Notion AI add-on in Settings & members > Plans. The add-on costs $4/month extra on top of any Notion plan. If it shows as purchased but still does not appear, log out of Notion completely (close the browser tab), wait 30 seconds, and log back in. If the issue persists, disable and re-enable Notion AI in your workspace settings.

---

Problem

Play.ht audio generation fails or produces low-quality output. Solution: First, verify you have an active paid subscription in Account > Billing. Check your monthly word count: if you exceed your plan's limit, generation stops until the next billing cycle. For quality issues, try a different voice: navigate to the Text to Speech editor, test 2–3 voice options from the dropdown, and listen to previews before generating. Avoid special characters, emojis, or very long paragraphs (split into multiple requests if needed).

---

Problem

Zapier/Make integration stops working after a few days or suddenly fails. Solution: Zapier and Make integrations can break if API keys expire or credentials become invalid. Navigate to your Zap/Scenario, click the module that failed (typically highlighted in red), and re-authenticate by clicking the Reconnect button. Re-enter API keys if prompted. Test the Zap by clicking Test before re-enabling automation. If the error persists, check the tool's status page (e.g., https://status.openai.com for OpenAI) to see if there is an ongoing outage.

---

Problem

Generated documentation is outdated or has duplicate content across wiki pages. Solution: Establish a documentation update schedule (e.g., weekly or on each product release). Use Notion's Synced blocks feature to prevent duplication: when you update content in one block, all synced copies update automatically. For large wikis, create a Notion database view sorted by Last edited date to quickly identify stale pages. Re-run the Cody + o1 + Claude workflow whenever major code changes occur, then manually merge new content with existing documentation in Notion using Claude's diff-aware editing.

---

Use your browser's print dialog to save as PDF (Ctrl+P / Cmd+P)