Skip to content

Latest commit

 

History

History
139 lines (87 loc) · 6.22 KB

File metadata and controls

139 lines (87 loc) · 6.22 KB

Native Chat

Nexus includes a full chat interface inside Obsidian — no need to switch to an external app.


Getting Started

  1. Configure a provider in Settings → Nexus → Providers
  2. Open chat via the ribbon icon or command palette (Nexus: Open Nexus Chat)
  3. Start typing — responses stream in real time

For voice and generated-media defaults, also review Settings → Nexus → Defaults.


Suggesters

Type special characters to trigger context-aware suggestions:

Trigger What It Does
/ Tool hints — browse and insert available tools
@ Custom prompts — invoke saved prompts
[[ Note links — reference vault notes inline
# Workspace data — pull in workspace context

Tool Calls

When the AI uses tools during a conversation, you see them as collapsible panels with live streaming results. Each tool call shows the agent, tool name, parameters, and output.


Voice And Media Defaults

Open Settings → Nexus → Defaults to configure the built-in voice and media surfaces:

  • Voice input sets the transcription provider/model used for chat microphone input and audio ingestion
  • Read aloud sets the speech provider/model/voice used when Nexus reads a note or selection aloud
  • Live voice sets the realtime provider/model/voice used by the chat composer voice session
  • Video sets the default provider/model, aspect ratio, and resolution used by generateVideo

The exact options depend on which providers and apps you have enabled.


Live Voice

Use the live voice button in the chat composer to start a realtime voice session inside the current conversation.

  • You must already have a conversation selected or created
  • User and assistant transcripts are appended back into the chat thread, so the voice exchange becomes part of the conversation history
  • The session uses the provider/model/voice selected under Settings → Nexus → Defaults → Voice → Live voice

Live voice comes in two shapes, and the model you pick decides which one you get:

Shape Models How it works
Native agent OpenAI GPT Realtime (WebRTC), Gemini Flash Live (WebSocket) The voice provider owns the whole conversation — it listens, thinks, and speaks
Composed pipeline OpenAI GPT Live Transcribe, AssemblyAI Universal 3.5 Pro Realtime The provider only transcribes. Your reply comes from your normal Nexus chat model and tools, then the configured speech model speaks it

Composed pipelines are the option to pick when you want live voice to use your actual chat model rather than whatever the voice provider hosts. For those models the voice dropdown reads "Uses speech default" and is disabled, since the voice comes from your speech model. Talking over a reply interrupts it — the answer you spoke over is discarded rather than spoken.

If a selected provider is not enabled and configured, Nexus shows an availability error instead of starting the session.


Read Aloud

Nexus can read either the active note or the current selection aloud.

You can start it from:

  • The command palette: Read note aloud, Read selection aloud, Stop read aloud
  • The editor context menu for selected text
  • The file context menu for Markdown notes

When the prompt offers Save & read, Nexus plays the audio and also writes a single audio file under your configured storage root and audio subfolder, then inserts a ![[...]] embed back into the note. Whole-note embeds go at the top of the note body; selection embeds are inserted immediately after the selected text.

Use Settings → Nexus → Defaults → Voice → Read aloud to choose the speech provider/model/voice, and Saved audio subfolder to choose where the generated files land.


Generated Media

Native chat also exposes built-in prompt tools for media generation when compatible backends are configured:

  • generateAudio creates spoken audio files directly in your vault using the configured Voice defaults or an explicit speech provider/model/voice
  • generateVideo creates MP4 files in your vault using Google or OpenRouter video models
  • checkGeneratedArtifact resumes a timed-out media job and saves the completed output to the requested vault path

generateVideo can return an in-progress result when the provider keeps rendering after the tool timeout. In that case, call checkGeneratedArtifact with the returned job ID instead of starting over.

See Provider setup for which providers unlock these tools.


Conversation Branching

Branch any conversation to explore alternative directions without losing the original thread. Branches are stored as linked conversations with parent metadata.


Providers

Configure providers in Settings → Nexus → Providers. All configured models appear in the chat model selector.

Provider Auth Notes
Anthropic API key sk-ant-...
OpenAI API key sk-proj-...
Google AI API key AIza...
Mistral API key msak_...
Groq API key gsk_...
OpenRouter API key or OAuth sk-or-... or sign in
Requesty API key req_...
Perplexity API key pplx-...
Ollama None Local, requires Ollama running
LM Studio None Local, requires LM Studio running
Claude Code Local CLI Must be installed and signed in on your computer first; no API key needed
Antigravity CLI Local CLI Install the Antigravity CLI, then run agy once to complete Google sign-in; no API key needed. Text-completion only (no tool calling)
GitHub Copilot OAuth device flow Requires active Copilot subscription; sign in via code in modal
Codex (ChatGPT) OAuth Requires ChatGPT Plus/Pro; sign in via browser redirect

See Provider setup for connection instructions for API key, local, CLI, and OAuth-backed providers.


Model Selection

Switch between any configured provider and model mid-conversation.


Subagents

The chat can spawn subagent conversations — branched LLM calls that handle tool continuations autonomously, then report results back to the main thread.