AI Chat
Interactive AI assistant for features and products with real-time collaboration and direct action capabilities.
AI Chat
AI Chat brings an intelligent assistant directly into your product and feature workflows. Instead of switching between tools, you can discuss requirements, request changes, and trigger actions — all through a natural conversation interface with real-time updates.
Why AI Chat?
Traditional project management tools create a gap between discussion and action. Teams discuss changes in one tool, then manually update statuses, create tasks, and modify artifacts in another.
Edsger's AI Chat closes this gap:
- Context-aware: The AI knows everything about your feature — status, workflow, user stories, test cases, technical design
- Actionable: The AI can directly update statuses, modify user stories, trigger phase re-runs, manage checklists, and create tasks
- Persistent: Conversations are saved and the AI remembers context across messages using session resumption
- Real-time: Messages appear instantly via Supabase Realtime — no polling or refreshing
- Team-wide: All team members see the same conversation and can collaborate with the AI together
Two Scopes of Chat
Feature Chat
Feature Chat is scoped to a single feature and has full access to modify it.
Context the AI knows:
- Feature name, description, status, and execution mode
- All workflow phases with their status (pending/completed/skipped)
- All user stories with titles, descriptions, and statuses
- All test cases with names, descriptions, and critical flags
- Recent chat history (up to 30 messages)
Actions the AI can take:
| Action | Description |
|---|---|
| Update feature status | Change status (e.g., back to ready_for_ai, backlog) |
| Update execution mode | Switch between full_pipeline, only_technical_design, from_code_implementation, etc. |
| Update workflow | Reset phases to pending, skip phases, or reorder |
| Manage user stories | Create, update, or delete user stories |
| Manage test cases | Create, update, or delete test cases (including is_critical flag) |
| Manage checklists | Create, update, or delete checklists and checklist items |
| Trigger phase re-run | Reset a specific workflow phase back to pending |
| Create tasks | Assign tasks to team members or AI |
| Read/write code | Browse and edit source code in the feature's repository |
Feature Chat also has access to Claude Code's built-in file tools (Read, Write, Edit, Bash, Glob, Grep), so the AI can directly read and modify source code when the feature has a connected repository.
Product Chat
Product Chat is scoped to a product and operates at a higher level.
Context the AI knows:
- Product name and description
- All features with names, statuses, and descriptions
- Team members with names, emails, and roles
- Recent chat history (up to 20 messages)
Actions the AI can take:
| Action | Description |
|---|---|
| List features | View all features, optionally filtered by status |
| Create features | Create new features for the product |
| Get feature details | Drill into any feature's full context |
| Get product overview | Summary with feature counts by status and team info |
| Manage checklists | Create, update, or delete product-level checklists |
| Create tasks | Assign tasks to team members or AI |
Product Chat does not have code access or feature-level mutation tools — it's designed for planning and oversight.
How It Works
Sending Messages
- Open a feature or product page
- Click the chat button (bottom-right floating icon) or navigate to the Chat tab
- Type your message and press Enter
- The AI processes your message and responds, potentially taking actions along the way
Interactive Options
The AI can present clickable option buttons when suggesting next steps. For example, after a phase completes, it might offer:
- "Continue to Technical Design"
- "Review User Stories First"
- "Re-run Feature Analysis with Feedback"
Clicking an option sends it as your message, continuing the conversation naturally.
Phase Completion Advisor
When a workflow phase finishes, the AI automatically:
- Posts a system message: "Phase X completed"
- Analyzes what was accomplished, including specific details (e.g., "8 user stories generated, 3 involve complex auth logic")
- Presents 2-4 actionable next steps as clickable options
This helps your team decide the best next action without having to remember the full workflow.
Session Persistence
The AI maintains conversation context across messages using session resumption:
- First message: Full context is loaded (feature/product state, workflow, user stories, test cases, recent history)
- Subsequent messages: Only the new message is sent — the AI already has the full conversation context from the session
- Session reset: If a session becomes corrupted, it automatically resets and reloads full context
This means the AI naturally remembers everything discussed in the channel without re-sending context every time.
Real-Time Updates
Chat uses Supabase Realtime for instant message delivery:
- New messages appear immediately for all team members
- The "AI is thinking..." indicator shows while the AI processes a message
- System messages (phase completions, errors) appear in real-time
- Unread message counts update on the floating chat button
User Interface
Floating Panel
A slide-in panel triggered by the circular chat button in the bottom-right corner. Shows an unread count badge. Ideal for quick interactions without leaving your current view.
Tab View
A full-height embedded panel under the "Chat" tab in the feature or product detail page. Includes a "Copy link" button for sharing the direct URL (?tab=chat).
Message Display
- Human messages: Right-aligned with user avatar
- AI messages: Left-aligned with bot avatar, rendered as full Markdown with a copy button
- System messages: Centered muted text (e.g., "Phase completed"). Consecutive identical system messages are collapsed with a repeat count (e.g., "x12")
- Option messages: Prompt text followed by clickable pill-shaped buttons
Pagination
Messages load in batches of 50. A "Load older messages" button appears when more history is available. The system automatically skips through consecutive identical system messages so you can reach real content faster.
Examples
Modifying User Stories via Chat
You: "Add a user story for admin users being able to export reports as CSV"
AI: I'll create that user story for you. [Creates user story: "As an admin user, I want to export reports as CSV so that I can analyze data in spreadsheets"] Done! I've added the user story with draft status.
Triggering a Phase Re-run
You: "The technical design needs to use PostgreSQL instead of MongoDB. Can you re-run it?"
AI: I'll reset the technical design phase. You may also want to add feedback so the AI knows about the PostgreSQL requirement. [Resets technical_design phase to pending] The technical design phase has been reset. When the workflow picks it up, it will regenerate the design.
Creating Tasks from Chat
You: "Create a task for @jane to review the API authentication design"
AI: [Creates task: "Review API authentication design", assigned to Jane, priority: medium] Done! I've created the task and assigned it to Jane. She'll see it in her task list.
Best Practices
- Use Feature Chat for specific changes: Update stories, test cases, trigger re-runs — the AI has full feature context
- Use Product Chat for planning: Create features, get overviews, discuss priorities across the product
- Be specific: "Add a test case for login with invalid password" works better than "add some tests"
- Use options: When the AI presents options, clicking them is faster than typing and ensures the AI understands your intent
- Combine with feedback: For complex changes, add formal feedback (which persists across AI re-runs) rather than relying on chat alone