AI panel
The AI panel is a dockable assistant that shares your workspace context. Toggle
it any time with
Providers — or none at all
Section titled “Providers — or none at all”- 12+ cloud providers — OpenAI, Anthropic, Google, Groq, xAI, Cerebras, DeepSeek, Mistral, OpenRouter, Hugging Face Inference API, and any OpenAI-compatible endpoint.
- Fully offline — LM Studio, MLX, and Ollama, with no API key required.
Configure them under Settings → AI. See AI providers for the full list and setup. Keys live in the OS keychain — never on disk.
Agents and tools
Section titled “Agents and tools”- Multi-agent and sub-agents with tool approval flows.
- Agent queue — queue follow-up tasks while an agent runs; they execute in order. Stop queued or running tasks from the Activity panel.
- File, shell, search, and plan tools — every one requires explicit approval before it executes.
- Slash commands and skills, plus prompt templates with
{VAR}placeholder substitution. - Semantic search — embed and search codebase symbols and docs so the AI retrieves the right context.
- Voice input — Whisper transcription via OpenAI.
- Auto-compaction keeps long conversations within the context window.
Transparency
Section titled “Transparency”- AI context inspector — a panel that shows exactly what context is sent to the model.
- Private terminals — the AI cannot read their scrollback.
- Project memory — drop a
NEXIS.mdin your project root and the AI remembers your conventions across sessions.
Terminal ↔ AI
Section titled “Terminal ↔ AI”- Select terminal output and click Ask Nexis or Explain to send it straight to the chat.
- Inline explain — select any terminal output or code and click Explain; the answer appears in a mini window without disrupting your flow.
- Explain commit — click a commit in git history for a plain-English breakdown of what changed and why.
The panel is dockable — resize, float, and snap it back to the dock.