AI panel

The AI panel is a dockable assistant that shares your workspace context. Toggle it any time with ctrl + icmd + ictrl + i.

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 → Models. See AI providers for the full list and setup. Keys are saved through the OS keychain, not in Nexis settings files.

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.
Tool strip — the AI window holds Chat, Queue, Refactor, Templates and Review. Handing work to the agent is a tab switch in the same window, and Chat stays mounted underneath, so a streaming reply survives a look at the queue.
File, shell, search, and plan tools — governed by per-tool prompt, deny, allow, or read-only auto-approval policies.
Custom TypeScript tools — scaffold workspace-specific agent capabilities without modifying Nexis itself.
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.

The orb

The new-chat card shows an animated orb, and while the agent works the “Thinking…” spinner becomes a small orb that follows the whole run. Pick one of 26 styles under Settings → AI → Orb, with a live preview and gallery. Ten styles follow the theme’s colours (marked with a Theme chip); sixteen keep a signature palette.

Nexis — Settings → AI → Orb — Settings → AI → Orb: a live preview of each state and the gallery of 26 styles.

stateDiagram-v2
  [*] --> Idle: new chat
  Idle --> Thinking: run starts
  Thinking --> Thinking: waiting on a tool approval
  Thinking --> Speaking: answer streams
  Speaking --> Thinking: next tool call
  Speaking --> Idle: run ends

States ease into each other rather than switching. Orbs are original WebGL2 shaders, so they run on every platform Nexis ships on. Without WebGL2 the orb falls back to a CSS disc, reduced motion draws one settled frame, and loops stop while the window is hidden or the orb is scrolled away.

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.md in your project root and the AI remembers your conventions across sessions.
Git-backed checkpoints — revert one agent turn without discarding other worktree changes.

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.