AI Chat

Lisa's core interface is conversational. You talk to her in plain language, and she turns that conversation into structured project artifacts — PRDs, issues, milestones, specs. No forms to fill, no fields to click through. Just describe what you want to build.

But this isn't a generic chatbot. Lisa has context. She knows your codebase, your existing projects, your open issues, and your team's templates. Every response is grounded in what's actually happening in your product — not hallucinated from thin air.

What Lisa knows about

When you start a conversation, Lisa already has access to:

• Your repositories — cloned and analyzed, so she can reference actual files, APIs, and architecture patterns when planning.
• Your projects — current scope, custom instructions, and any artifacts (PRDs, specs, notes) attached to them.
• Your issues — existing work items, their status, dependencies, and hierarchies.
• Your milestones — timeline and progress context for scoping new work.
• Your templates — your team's standard formats for PRDs, specs, and issues, so output matches your expectations.

This is what makes Lisa code-aware. When she suggests breaking a feature into issues, she's referencing your actual codebase — not guessing at what your tech stack looks like.

How to talk to Lisa

Be direct. Lisa works best when you tell her what you want to accomplish, not how to accomplish it. She'll ask clarifying questions when she needs more context.

Tips for better results

• Not sure what to ask? Just ask Lisa. She's an AI — you can say "I want to plan a new feature but I'm not sure where to start, help me think through this" and she'll help you shape the right questions. There's no wrong way to begin a conversation.
• Be specific about outcomes. "Create issues" is fine. "Create issues for the payment integration with acceptance criteria and dependency mapping" is better.
• Reference existing context. Mention projects, milestones, or repos by name — Lisa uses that to scope her responses.
• Build on the conversation. Lisa remembers everything in the current session. You don't need to repeat yourself — refine, redirect, and iterate.
• Start with Discovery for big features. When you have a complex idea that needs structure, ask Lisa to start a discovery session to go from vague concept to executable plan.

Chat presets

Presets are quick-start prompts that appear based on what you're working on. Instead of starting from a blank message, they give you a jumping-off point:

• Working on a project? You'll see presets like "Start discovery", "Review roadmap", "Create milestone", "Summarize status", and "Identify priorities".
• Looking at an issue? Presets include "Write technical spec", "Add acceptance criteria", "Break into subtasks", "Estimate effort", and "Identify dependencies".
• Inside a milestone? You'll get "Review progress", "Identify blockers", "Create status report", "Adjust timeline", and "Plan next steps".

Presets aren't magic — they're pre-configured prompts that steer Lisa toward a specific type of output. You can always type your own prompt instead.

Model selection

In BYOK (Bring Your Own Key) mode, Lisa supports multiple AI providers so you can choose the model that fits your workflow:

• Anthropic — Claude Opus 4.6, Claude Sonnet 4.5, Claude Haiku 4.5. Claude excels at structured planning, long-form documents, and nuanced reasoning.
• OpenAI — GPT-5.2, GPT-5 Mini, GPT-5 Nano. Strong at code analysis and fast iteration.
• Google AI — Gemini 3 Pro, Gemini 3 Flash. Good for large-context tasks and quick responses.

Switch models from the model selector in the chat interface. Each conversation remembers your model choice, and you can switch mid-conversation if you want to compare outputs.

In SaaS mode, Lisa uses an optimized default model — no configuration needed.

How chat works under the hood

When you send your first message, Lisa creates a new session and starts streaming responses in real time. You'll see her thinking and working — when she calls tools (like creating an issue or analyzing a repo), each action appears as an expandable card showing what's happening.

If Lisa needs to do something potentially impactful — like creating multiple issues or modifying existing work — she'll ask for your permission before proceeding.

Every conversation within a project maintains its own session with full history. This means you can run multiple planning threads in parallel — one for the auth feature, another for the notification system — without context bleeding between them. Switch between sessions from the project view to pick up where you left off.