Home23

Settings

Control Surface
← Back to Dashboard
SETTINGS

Models

Mixed

Configure the default chat model for agents. Advanced routing and house-wide model catalogs are still available below, but they stay out of the main path.

Mixed Scope
Agent-scoped tabs target the selected agent. Providers, Skills, Vibe, Tiles, and System stay house-wide.

Providers & API Keys

OAuth is the preferred way to authenticate with Anthropic and ChatGPT. It's free with a Claude Max plan or ChatGPT Plus/Pro subscription and refreshes automatically. Use API keys below for all other providers or as a fallback.

OAuth Sign-in recommended

Anthropic (claude.ai)
Sign in with your Claude account for Sonnet / Opus access.
checking…
OpenAI Codex (ChatGPT)
Sign in with your ChatGPT account for GPT-5 / Codex access without an API key.
checking…

API Keys

Stored in config/secrets.yaml. Use these for providers that don't offer OAuth, or as a fallback.

Agents

Roster and runtime controls for each Home23 agent.

Worker Library

Install reusable specialists that can diagnose, verify, summarize, and report back without running a standalone Home23 engine. This is where the house learns what each worker is for.

What This Unlocks

Install Specialist

Installed Specialists

Available Worker Packs

Try A Worker

Recent Proof Trail

Terminal Use

node cli/home23.js worker list node cli/home23.js worker create systems --template systems --owner jerry node cli/home23.js worker run systems "inspect the current live problem and produce a receipt"

Models

Pick the normal chat model for the selected agent. Advanced routing and house-wide model catalogs are still available below, but they stay out of the main path.

Chat Default

The model this agent uses for normal chat unless a conversation explicitly picks another model.

Pulse Voice

Optional model override for the selected agent's dashboard remarks. Leave it aligned with chat unless the pulse needs a different voice model.

Voice Prompt Editor Dashboard remark personality for this agent.

How the dashboard voice talks. Tune this when the pulse remarks need a different personality.

Advanced Cognitive Routing Internal engine slots, primary models, and fallbacks.

Per-slot routing for the thinking engine. Each row is one internal model call path, with optional fallbacks for errors or rate limits.

House-Wide Catalog and Aliases Shared provider model lists and shortcut names.

Available Models per Provider

House-wide model catalog. These are the models shown in evobrew and shared pickers across Home23.

Aliases

House-wide shortcuts that any agent can use when resolving model names.

AliasProviderModel

Query

Defaults for the selected agent's Query tab. These seed the dropdowns when that dashboard loads; users can still override per-query.

Default Query

PGS (Partitioned Graph Synthesis)

Document Feeder

Configure what documents get ingested into your agent's brain and how they're processed. These settings target the selected agent and save to its instances/<agent>/config.yaml, with configs/base-engine.yaml acting only as the fallback default. Some changes hot-apply; others require that agent's engine to restart.

Live Status

—
—
—
—
—
—
—

Paths & Patterns

The orchestrator always watches the drop zone and the agent's workspace directory. Add additional paths here to ingest files from elsewhere on your machine.

Frequency & Batching

Drives when pending documents are flushed from the ingestion queue to the live brain. Interactive uploads always trigger a debounced fast flush regardless of these settings.

Document Compiler

The compiler runs raw documents through an LLM to extract concepts, relationships, and insights before they enter your agent's brain. Disabling means documents go in as raw chunks — lower quality, more noise.

Binary Format Converter

Uses MarkItDown (Python) to convert PDFs, DOCX, images, audio, and other binary formats to markdown before ingestion. Requires pip install markitdown. Vision model needs OpenAI API access.

—

Drop Zone

Drag and drop files to ingest them into your agent's brain. Files land in the watched drop zone directory and are auto-processed within a second. Max 100MB per file, 20 files per upload.

⬇
Drop files here or click to browse
Any format — text files go direct, binaries via the converter

Skills

Shared skills are host-wide capabilities for all Home23 agents on this machine. Public skill settings live in config/home.yaml; credentials live in config/secrets.yaml.

Skill Catalog

Use this to see what the shared skill system currently exposes, how healthy each skill is, and which ones need credentials.

X Research

Read-only X/Twitter research uses its own X API bearer token. This is stored host-wide so any Home23 agent can use the skill without per-user shell setup.

No token configured
0
Cheap pulse checks first unless a prompt explicitly asks for deeper research.
Writes research summaries into the workspace report path automatically.

Vibe

Controls the dashboard Vibe tile — CHAOS MODE image generation, archive rotation, and optional dream augmentation that pulls symbolic motifs from the agent's most recent dreams.

Generation & Rotation

External Sources

Additional directories of images to mix into the Vibe cycle. One absolute path per line (use ~/ for your home directory). Images from these folders are displayed alongside CHAOS MODE generations — sorted by file modification time, newest first.

Dream Augmentation

When enabled, the vibe service reads the N most recent dreams from the agent's brain and extracts symbolic motifs. Motifs are woven into CHAOS MODE prompts as texture and atmosphere, never as the literal subject.

Tiles

Compose the Home dashboard without code changes. Layout controls the full Home tab. Custom tiles use reusable connections stored in config/secrets.yaml; tile definitions live in config/home.yaml.

Layout

Drag rows to reorder the Home tab. Toggle visibility and set each tile's width.

Tiles

Create custom weather, sauna, or advanced HTTP/JSON tiles.

Tile Builder

Connections

Reusable secret-backed connections for all custom tiles.

Connection Builder

Agency — Autonomous Actions

The allow-list below is house-wide. The audit trails at the bottom show what the selected agent actually tried and executed.

Global

Actions

Toggle actions on/off, flip into dry-run (logs intent only, no side effect), adjust per-action rate limits. Changes hot-apply — no restart.

Integrations

Recent Actions

Live audit trail for the selected agent. Intent is logged before the handler runs; outcome after. Refreshes every 5s.

Rejected Requests

Actions the selected agent tried but weren't in the allow-list. Worth watching — these are requests for new capabilities.

System

Ports, embeddings, chat defaults, and maintenance actions.

Ports

Chat Defaults

Embeddings

Changing the embedding model or dimensions requires re-ingesting all documents. Existing brain embeddings become incompatible with the new model.

Maintenance