ComfyChat CONDUCTOR

LOCAL-FIRST · CHAT-FIRST

Your AI fleet,
one conversation
away.

ComfyChat brings local models, on-device intelligence, image engines, and the machines behind them into one calm iPhone and iPad workspace.

Follow the flow Private source
GitHub access is required to view this private repository.
COMFYCHAT · SESSION3 NODES ONLINE
Create a cinematic night-city image, then ask two models to compare the composition.
LOCAL · QWENI’ll sketch the visual structure and hand the approved prompt to the image tool.
generate_imageComfyUI · checkpoint ready
RUNNING
Message your fleet…↑
LM STUDIOREADY
COMFYUI2 JOBS
ON DEVICEAVAILABLE
NO HOSTED BACKENDNO APP ACCOUNTNO TELEMETRYLAN + TAILSCALE

ONE COCKPIT, MANY ENGINES

The stack you already own becomes one creative surface.

Nodes remain visible without taking over the experience. Reachability, model identity, engine health, queues, memory, checkpoints, and failures stay attached to the work they affect.

01

Talk to models where they run

Use Apple Foundation Models when available and configured LM Studio nodes across your private network.

CHAT · MODEL IDENTITY · STREAMING
02

Create from the conversation

Send image work through MCP-shaped tools backed by ComfyUI and compatible engines, then keep results in chat and the image library.

TOOLS · BATCH RUNS · IMAGE LIBRARY
03

See the machine behind the answer

Keep node, engine, queue, checkpoint, and run status understandable before, during, and after generation.

NODES · HEALTH · OBSERVABILITY

NATURAL LANGUAGE + TRUSTWORTHY TOOLS

Ask naturally.
Review precisely.

A model can request a tool, or you can launch a deterministic image flow directly. Either path is validated, tracked, and returned to the conversation.

  1. 01
    Choose the participantsSelect one or more local or on-device chat models.
  2. 02
    Describe the outcomeKeep the request in the conversation instead of jumping between dashboards.
  3. 03
    Run the right engineValidate the tool call, node, checkpoint scope, and settings.
  4. 04
    Keep the evidenceReview labeled responses, progress, results, and failures in context.

REPOSITORY CHECKPOINT · AUGUST 2026

Built enough to use.
Validating enough to trust.

Source and automated tests prove the product foundation. Live fleet acceptance and public-release evidence remain separate gates—not implied completion.

BUILT

Chat, tools, and durable creative runs

Local and on-device chat, labeled multi-model sessions, MCP-shaped tool dispatch, ComfyUI generation and batch flows, recovery-aware run records, and a file-backed image library are implemented.

NOW · VALIDATION

Prove the real fleet and release path

Finish live single-node and two-node ComfyUI acceptance, confirm final repository gates, and close the public privacy, support, screenshot, and release-proof checklist.

NEXT

Make nodes and model modes legible

Deepen engine health, queues, memory, checkpoint status, safe controls, targeted model turns, round-robin rooms, and side-by-side comparison.

LATER

Bring the fleet onto the device

Physically validate managed GGUF, MLX, and Core ML artifacts, then add supported App Intents, Live Activities, widgets, and a native Mac surface.

ComfyChat app iconiPHONE · iPAD · APPLE SILICON MAC

LOCAL-FIRST BY DESIGN

Your models. Your machines. Your files.

ComfyChat is not a hosted AI service. It is a mobile command center for the local stack you already run, with on-device capability where supported and explicit routes to configured engines.

View source on GitHub