Olla Chat

Professional AI coding assistant for VS Code, built Ollama-first.

Olla Chat gives you Ask, Plan, and Agent workflows in one sidebar, with streaming responses, context-aware editing, human approvals, and local-model control.

Why Olla Chat

• Ollama-first local or remote model routing
• Multi-session conversations with mode + model per chat
• Selection-aware editing for real code writing workflows
• Human-in-the-loop safety for file-changing actions
• Clean, theme-aware interface designed for daily dev use

Core Features

• Modes: Ask, Plan, Agent
• Streaming UI: token streaming + timeline events
• Context controls: Selection, File, Open Files, Project
• Smart edits: selection replace flow with apply/undo
• Attachments: file + image upload support
• Vision model checks: warns when selected model is not vision-capable
• Model controls: picker + temperature at composer level
• Debug observability: detailed internal logs via Output channel

Quick Start

1. Install the extension.
2. Start Ollama (or point to your remote Ollama endpoint).
3. Open Olla Chat from the Activity Bar.
4. Choose a model and mode, then send a prompt.

Default endpoint:

http://localhost:11434

How To Use

1. Choose a mode

• Ask: direct Q&A, explanations, focused answers
• Plan: structured step-by-step planning before action
• Agent: autonomous execution with approval gates for risky operations

2. Control context

Use scope chips to decide what the model sees:

• Selection for selected text/code
• File for active editor file
• Open Files for broader current-work context
• Project for repository-level understanding

3. Edit from chat

• In Agent mode, selection-targeted edit responses can auto-apply.
• In Ask mode, you get an explicit apply confirmation.
• Use Undo from chat to revert applied selection replacements.

4. Work with attachments

• Attach files/images from the composer.
• For image analysis, select a vision-capable model.

Configuration

Settings prefix: olla-chat.*

• ollamaUrl - Ollama base URL
• ollamaModel - default model
• temperature - generation temperature (0 to 2)
• defaultMode - ask | plan | agent
• approvalPolicy - human_gated | auto_safe
• contextPolicy - auto_light | manual_only | always_project
• maxContextFiles - max files sampled for project context
• debugLogs - verbose logs in Olla Chat Debug

Troubleshooting

• Model not found: run ollama list, then ollama pull <model>.
• Image error on vision request: choose a model marked vision-capable.
• No response stream: verify ollamaUrl, model availability, and logs (Olla Chat Debug).