Argonaut Hub

The Argonaut Hub lives under /admin/argonaut/* and provides the management interface for the Argonaut AI agent. It spans six pages: Dashboard, Chat, Profile, Profiles, Tasks, and Writer. Argonaut is the primary AI persona for Arcturus-Prime — it handles conversational interactions, content generation, RAG-augmented queries, and writing assistance. The hub gives administrators control over every aspect of the agent’s configuration, state, and output.

Dashboard (/admin/argonaut)

The Argonaut dashboard is the status and control center for the agent. It surfaces three primary areas: agent status, model configuration, and RAG pipeline controls.

Agent Status

The dashboard polls /api/argonaut/status on load and displays the current state of the Argonaut agent: whether the agent is online, what model is currently loaded, the active profile chain, token usage for the current billing period, and the last interaction timestamp. A status indicator shows green (healthy), yellow (degraded — high latency or model loading), or red (offline or errored).

Model Selection

The model selector pulls available models from /api/argonaut/models and displays them grouped by provider. Changing the model here sets the default model for all Argonaut interactions until changed again. The selector shows each model’s context window size, cost tier, and a brief capability note. The currently active model is highlighted with a badge.

RAG Pipeline Controls

The RAG section provides direct control over the ingestion pipeline through /api/argonaut/ingest. Three actions are available:

Full Re-index — rebuilds the entire embedding index from all content collections (posts, journal, docs, projects, learn). This is a heavy operation that processes every content file, chunks it, generates embeddings, and writes the updated index to /public/embeddings-index.json.
Incremental Update — re-indexes only content files that have changed since the last index build. Compares file modification timestamps against the index metadata.
Index Status — displays the current index statistics: total documents, total chunks, file size, last build timestamp, and per-collection breakdowns.

Chat (/admin/argonaut/chat)

The Argonaut chat page is a full conversational interface with the AI agent. It goes beyond the basic /admin/chat by integrating Argonaut-specific features: profile-aware system prompts, RAG-augmented responses, and backend selection.

Conversation Interface

The main area is a standard chat layout with message bubbles, a text input with send button, and streaming response rendering. Messages support Markdown formatting, code blocks with syntax highlighting, and inline images.

History Sidebar

A collapsible sidebar on the left lists all past conversations sorted by last activity. Each entry shows the conversation title (auto-generated or manually set), message count, and timestamp. Conversations can be renamed, pinned, or deleted from the sidebar. The sidebar also has a “New Conversation” button that starts a fresh thread.

Model Selector

A dropdown in the chat header allows overriding the default model for the current conversation. The selector shows all available models across all providers. The selected model persists per conversation, so switching conversations also switches the active model.

Backend Toggle

A toggle switches between two backend modes:

Argonaut — routes messages through the Argonaut agent pipeline, which applies the full profile chain (BOOT, IDENTITY, SOUL, etc.) as system context. Responses reflect the Argonaut persona.
Claw — routes messages through the OpenClaw backend, which uses a different system prompt and tool chain. Useful for tasks that need OpenClaw’s skill system rather than Argonaut’s persona.

RAG Mode Toggle

A second toggle controls how the RAG pipeline is applied to queries:

Full — every user message triggers a RAG retrieval pass. Relevant content chunks from the embedding index are injected into the context before the model sees the message. This provides maximum context but increases token usage.
Safe — RAG retrieval is only triggered when the query appears to reference Arcturus-Prime content, infrastructure, or documentation. A lightweight classifier determines relevance before running the retrieval pass. This reduces unnecessary context injection for general questions.

Voice Activity Detection

The chat supports voice input via the browser’s Web Speech API. A microphone button toggles Voice Activity Detection (VAD), which listens for speech and automatically transcribes it into the text input. The transcription runs client-side using the browser’s built-in speech recognition. VAD stops automatically after a configurable silence duration (default: 2 seconds).

Sources Panel

When RAG mode is active, a collapsible sources panel appears below each AI response. It lists the content chunks that were retrieved and injected as context, showing:

Source document — the title and collection of the source content
Chunk preview — the first 200 characters of the retrieved chunk
Confidence score — the similarity score (0.0 to 1.0) between the query embedding and the chunk embedding. Scores above 0.8 indicate high relevance.

Profile (/admin/argonaut/profile)

The profile page is a read-only viewer for the currently active Argonaut agent profile. It displays:

Agent name — the configured agent identifier
Active status — whether the agent is currently loaded and responsive
Profile chain — the ordered list of profile files that compose the agent’s system prompt
File locations — absolute paths to each profile file on disk
Merged system prompt — the final concatenated system prompt that the agent receives, assembled from all profile files in load order

This page does not support editing. Profile file contents are edited through the Profiles page. The Profile page exists for quick reference and debugging — verifying what the agent is actually seeing as its system prompt.

Profiles (/admin/argonaut/profiles)

The profiles page is a file browser and editor for Argonaut profile files. Profile files are loaded in a strict order that defines the agent’s persona, knowledge, and behavior.

Profile Load Order

Profiles are loaded and concatenated in this order:

BOOT — base system instructions, safety constraints, and operational boundaries
IDENTITY — who the agent is: name, role, relationship to Arcturus-Prime
SOUL — personality, voice characteristics, tone, humor style
AGENTS — multi-agent coordination rules and inter-agent communication protocols
USER — user-specific context: who the user is, their preferences, their role
TOOLS — available tool definitions, API access, and capability descriptions
HEARTBEAT — dynamic state injected at runtime: current time, recent activity, system health

File Browser

The left panel lists all profile files with their load order position, filename, and file size. Clicking a file loads its contents in the editor panel on the right.

Editor

The right panel provides a text editor for modifying profile file contents. Changes are saved through /api/argonaut/profiles using PUT requests. The API supports both GET (retrieve file contents) and PUT (update file contents) operations. A diff viewer shows pending changes before saving. The editor includes syntax highlighting for Markdown and YAML frontmatter that may appear in profile files.

Tasks (/admin/argonaut/tasks)

The tasks page provides a task management interface for tracking work items assigned to or generated by the Argonaut agent.

Task Properties

Each task has:

Title — short description of the work item
Status — one of: pending, in-progress, completed
Priority — low, medium, high, urgent
Lane — categorization for grouping: content, infrastructure, security, ai, general
Description — detailed notes and context
Created/Updated timestamps

Persistence

Tasks are persisted to localStorage in the browser. There is no server-side task storage — this is a lightweight personal tracking tool, not a shared project management system. Tasks survive page reloads and browser restarts but do not sync across devices.

Interface

The task list renders as a sortable, filterable table. Filters are available for status and lane. New tasks are created via a form at the top of the page with fields for title, description, priority, and lane. Status changes are made by clicking status badges on each task row, which cycles through the three states. Completed tasks can be hidden with a toggle or cleared in bulk.

Writer (/admin/argonaut/writer)

The writer page is a specialized blog content creation interface that combines a Markdown editor with AI chat assistance, voice scoring, and template management.

Split Pane Layout

The page renders in a two-column split pane:

Left pane (Editor) — a Markdown editor for writing the blog post. Supports syntax highlighting, live preview toggle, and standard editor shortcuts. The editor starts with a template or blank document.
Right pane (Chat) — an Argonaut chat interface scoped to writing assistance. The chat has context about the current editor content — it can see what you have written and provide feedback, suggestions, rewrites, and expansions. Asking “make the intro punchier” works because the chat knows the current document state.

Voice Scoring

A voice score widget in the editor toolbar shows the current document’s alignment with the Arcturus-Prime voice profile. The score is fetched from /api/argonaut/voice-score and ranges from 0 to 100. The widget updates when the user triggers a manual score check (not on every keystroke). The score breaks down into dimensions: tone, vocabulary, structure, technical depth, and personality. Low-scoring sections are highlighted in the editor with inline annotations suggesting improvements.

Template Selection

A template picker offers starting points for common content types: blog post, tutorial, deep dive, journal entry, project showcase, and learning guide. Each template includes pre-filled frontmatter and structural headings appropriate to the content type. Templates pull from the Arcturus-Prime content collection schemas.

Frontmatter Generation

A “Generate Frontmatter” button analyzes the current document content and produces suggested frontmatter: title, description, tags, category, and estimated read time. The generated frontmatter follows the posts collection schema and can be inserted at the top of the document with one click.

Sanitization Scoring

A sanitization check runs alongside voice scoring, scanning the document for unsanitized content: real IPs (10.42.0.x, 192.168.20.x), real hostnames (Altair-Link, Meridian-Host, Tarn-Host), and other identifying information that should be mapped through the Galactic Identity System before publishing. The check surfaces violations with line numbers and suggested replacements. This only applies to public-facing content — internal docs skip this check.

Search Integration

The writer integrates with /api/argonaut/search to let authors search existing Arcturus-Prime content while writing. This helps avoid duplicate topics, find related posts to link to, and maintain consistency with previously published material. Search results appear in a popover panel and can be inserted as Markdown links directly into the editor.