AI Chat

The chat view is the heart of Trident: two independent AI conversations running side by side, often with models from different providers, so you can compare takes, iterate on ideas, or divide the work without switching contexts.

Dual-Panel Layout

When you open a project, you see left and right chat panels. Each panel is a fully independent conversation with its own:

• Model selection from any configured provider connection
• Conversation history
• Document attachments
• Token usage tracking

The panels are resizable. Drag the divider to give more space to either side.

Choosing a Model

Each conversation has a model selector at the top. Click it to switch between any configured model. The model is locked to the conversation once you start chatting, so different conversations can use different models simultaneously.

You need at least one provider connection configured in Settings > Providers before you can chat.

Models from cloud connections are listed under their platform (Amazon Bedrock, Google Vertex AI, or Azure OpenAI), and the search box matches on model name, so you can find a model without knowing how its deployment or inference profile is named.

Reasoning Effort

Models that support extended thinking show an effort selector next to the model name, from Low through Max. Higher effort lets the model think longer before answering, which helps on harder problems and costs more. The setting is saved per conversation, and the selector is hidden for models that don’t offer the control.

Attaching Documents and Images

Below the message input, Attach Project Files lets you select documents and images from the project to send with your message. An attached document gives the model its full content, which it can reference or edit in its response. Attachments work across providers. You can attach a document originally drafted by one model to a conversation with another.

Images can be attached to any model that accepts visual input. If the selected model doesn’t, the image picker says so. For a cloud deployment with a custom name, setting its base model ID in Settings > Providers lets Trident identify what the model supports.

What the Model Can Do

Every model you chat with has access to the same set of project tools, regardless of provider:

• Read and edit attached documents: the model sees the full content and can rewrite, extend, or fix anything you’ve attached to the conversation, including documents that a different model originally wrote. This is how models end up collaborating on the same piece of work: draft with one, then attach and hand off to another for review or edits.
• Create new documents: ask the model to draft something and it lands in the project alongside your own documents, ready to attach to a later conversation with any model.
• Generate images: ask any model for an image and it opens a picker inline so you choose which of your connected image models to use, along with dimensions and quality. The finished image is saved to the project gallery.
• Read your workspace directory: if you set one when you created the project, the model can list, search, and read files inside it (read-only; it can’t write back to the workspace).

Conversations

Each panel can have multiple conversations. Click the conversation history sidebar to switch between them or create a new one. Conversations track:

• Title: generated from your first message by the conversation’s own model, so no prompt goes to a provider you didn’t pick for it. You can rename it at any time.
• Message count
• Which model was used
• Last message timestamp

Right-click a conversation to rename or delete it.

Streaming Responses

AI responses stream in real time as the model generates them. You’ll see a token usage indicator showing how much of the model’s context window has been used, along with any context caching information.

Initial Prompt

If your project has an initial prompt configured, it will be automatically sent to both chat panels the first time you open the project. This is useful for establishing context or giving the AI standing instructions.