Part of AbstractGateway v0.7.2 (terminal console v0.11.1) · part of AbstractCore v2.19.0 (terminal console v0.4.1)

AbstractGateway and AbstractCore consoles

The consoles are part of their packages, not packages of their own. AbstractGateway comes with a web console at /console and a terminal console, abstractgateway-console, for servers without a browser; both use the same gateway routes and the same admin rules. AbstractCore, used on its own as a library, has its own pair: the web console of abstractcore serve and abstractcore-console. Both terminal consoles are built on AbstractTUI.

Gateway web console Gateway terminal console AbstractCore consoles Console reference crates.io

# AbstractGateway web console: start the gateway, open the one-time link it prints
abstractgateway serve          # http://127.0.0.1:8080/console#claim=…
abstractgateway claim --open   # a fresh link (valid 10 minutes, this machine only)

# AbstractGateway terminal console: built by the installer, or
cargo install abstractgateway-console
abstractgateway-console --gateway-url http://127.0.0.1:8080 --token <admin token>

AbstractGateway · First run

A guide that sets up this machine

The first time you sign in, the gateway console opens a setup guide. Every step is optional, and the Setup button (top right, admins) reopens it later.

Step	What it shows
Welcome	This machine (memory, GPU), the data folder, the sign-in mode, and whether the gateway starts at login.
Local engines	The engines found here (Ollama, LM Studio, MLX, llama.cpp, …), installed or not, running or not, with an Install button that shows the exact command first.
Default model	Use recommended defaults for text, voice, images and video where this computer can run them, a Download button for each missing model, and the catalog filtered to models that fit.
Apps	Observer, Continuum, Code, Entity and the Flow Editor, then the desktop Assistant: Install, then Open.
Done	How to reopen the console, the Start at login switch, and the terminal console's commands.

The gateway web console's first-run guide, step 1 of 5, showing this machine's memory, GPU, data folder and sign-in mode — The first-run guide in the web console.

What "recommended" means on your computer

AbstractCore recommends one model per capability for the kind of computer the gateway runs on, and Use recommended defaults writes the starter set (text, speech output, images and video) where this computer can run it. Speech input and music are recommended too, but you set them yourself. For text on an Apple silicon Mac the recommendation follows memory (8 GB up to 24 GB: Qwen3.5 9B; 24 GB up to 128 GB: Qwen3.8 27B; from 128 GB: Qwen3.8 Flash-Next). When a model may not fit, the console says so instead of switching silently, and a capability this computer cannot run shows the reason. Fit notes assume macOS's default GPU memory limit (75% of unified memory).

This computer	Text and chat	Speech output (text to speech)	Speech input (speech to text)	Music generation	Image input (vision)	Image generation	Video generation
Apple silicon Mac, 8 GB	`Qwen3.5 9B` MLX (AbstractCore), Apple GPU (Metal) fits tightly	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	Not available ACE-Step 1.5 XL turbo (music) needs about 10.9 GiB of memory while it generates (estimated), and macOS's GPU memory limit on this Mac is about 6.0 GiB, about 4.0 GiB of it left for a model after working buffers; use a computer with more memory, or a cloud music backend (acemusic or elevenlabs-music, with its API key)	Read by the text model	Not available FLUX.2 [klein] 4B (8-bit) needs about 8.5 GiB of memory while it generates (estimated), and macOS's GPU memory limit on this Mac is about 6.0 GiB, about 4.0 GiB of it left for a model after working buffers; use a Mac with more unified memory, or a cloud image provider	Not available Wan2.2 TI2V 5B (text/image to video) needs about 16.6 GiB of memory while it generates (measured with AbstractVision with MLX-Gen at 832x480x121 for image-to-video, its larger task, text-to-video needing 16.3 GiB; the engine keeps the text encoder and VAE in memory; this engine's figure, not the model's minimum), and macOS's GPU memory limit on this Mac is about 6.0 GiB, about 4.0 GiB of it left for a model after working buffers; use an Apple silicon Mac with more unified memory, or an OpenAI-compatible video endpoint (AbstractVision OpenAI-compatible backend)
Apple silicon Mac, 16 GB	`Qwen3.5 9B` MLX (AbstractCore), Apple GPU (Metal)	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	Not available ACE-Step 1.5 XL turbo (music) needs about 10.9 GiB of memory while it generates (estimated), and macOS's GPU memory limit on this Mac is about 12.0 GiB, about 10.0 GiB of it left for a model after working buffers; use a computer with more memory, or a cloud music backend (acemusic or elevenlabs-music, with its API key)	Read by the text model	`FLUX.2 [klein] 4B (8-bit)` MLX-Gen (AbstractVision), Apple GPU (Metal) fits tightly	Not available Wan2.2 TI2V 5B (text/image to video) needs about 16.6 GiB of memory while it generates (measured with AbstractVision with MLX-Gen at 832x480x121 for image-to-video, its larger task, text-to-video needing 16.3 GiB; the engine keeps the text encoder and VAE in memory; this engine's figure, not the model's minimum), and macOS's GPU memory limit on this Mac is about 12.0 GiB, about 10.0 GiB of it left for a model after working buffers; use an Apple silicon Mac with more unified memory, or an OpenAI-compatible video endpoint (AbstractVision OpenAI-compatible backend)
Apple silicon Mac, 18 GB	`Qwen3.5 9B` MLX (AbstractCore), Apple GPU (Metal)	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	`ACE-Step 1.5 XL turbo (music)` ACE-Step on Diffusers and PyTorch (AbstractMusic), Apple GPU (MPS, bfloat16) fits tightly set it yourself	Read by the text model	`FLUX.2 [klein] 4B (8-bit)` MLX-Gen (AbstractVision), Apple GPU (Metal)	Not available Wan2.2 TI2V 5B (text/image to video) needs about 16.6 GiB of memory while it generates (measured with AbstractVision with MLX-Gen at 832x480x121 for image-to-video, its larger task, text-to-video needing 16.3 GiB; the engine keeps the text encoder and VAE in memory; this engine's figure, not the model's minimum), and macOS's GPU memory limit on this Mac is about 13.5 GiB, about 11.5 GiB of it left for a model after working buffers; use an Apple silicon Mac with more unified memory, or an OpenAI-compatible video endpoint (AbstractVision OpenAI-compatible backend)
Apple silicon Mac, 24 GB	`Qwen3.8 27B` MLX (AbstractCore), Apple GPU (Metal) fits tightly	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	`ACE-Step 1.5 XL turbo (music)` ACE-Step on Diffusers and PyTorch (AbstractMusic), Apple GPU (MPS, bfloat16) set it yourself	Read by the text model	`FLUX.2 [klein] 4B (8-bit)` MLX-Gen (AbstractVision), Apple GPU (Metal)	Not available Wan2.2 TI2V 5B (text/image to video) needs about 16.6 GiB of memory while it generates (measured with AbstractVision with MLX-Gen at 832x480x121 for image-to-video, its larger task, text-to-video needing 16.3 GiB; the engine keeps the text encoder and VAE in memory; this engine's figure, not the model's minimum), and macOS's GPU memory limit on this Mac is about 18.0 GiB, about 16.0 GiB of it left for a model after working buffers. It fits once macOS lets the GPU use 20 GiB: run `sudo sysctl iogpu.wired_limit_mb=20480` in a terminal (asks for your password; lasts until the Mac restarts), then load it.
Apple silicon Mac, 32 GB, 36 GB, 48 GB, 64 GB or 96 GB	`Qwen3.8 27B` MLX (AbstractCore), Apple GPU (Metal)	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	`ACE-Step 1.5 XL turbo (music)` ACE-Step on Diffusers and PyTorch (AbstractMusic), Apple GPU (MPS, bfloat16) set it yourself	Read by the text model	`FLUX.2 [klein] 4B (8-bit)` MLX-Gen (AbstractVision), Apple GPU (Metal)	`Wan2.2 TI2V 5B (text/image to video)` MLX-Gen (AbstractVision), Apple GPU (Metal)
Apple silicon Mac, 128 GB	`Qwen3.8 Flash-Next` MLX (AbstractCore), Apple GPU (Metal) raise GPU limit `sudo sysctl iogpu.wired_limit_mb=114688`	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	`ACE-Step 1.5 XL turbo (music)` ACE-Step on Diffusers and PyTorch (AbstractMusic), Apple GPU (MPS, bfloat16) set it yourself	Read by the text model	`FLUX.2 [klein] 4B (8-bit)` MLX-Gen (AbstractVision), Apple GPU (Metal)	`Wan2.2 TI2V 5B (text/image to video)` MLX-Gen (AbstractVision), Apple GPU (Metal)
Apple silicon Mac, 192 GB, 256 GB or 512 GB	`Qwen3.8 Flash-Next` MLX (AbstractCore), Apple GPU (Metal)	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	`ACE-Step 1.5 XL turbo (music)` ACE-Step on Diffusers and PyTorch (AbstractMusic), Apple GPU (MPS, bfloat16) set it yourself	Read by the text model	`FLUX.2 [klein] 4B (8-bit)` MLX-Gen (AbstractVision), Apple GPU (Metal)	`Wan2.2 TI2V 5B (text/image to video)` MLX-Gen (AbstractVision), Apple GPU (Metal)
Linux or Windows with an NVIDIA GPU	`Qwen3.5 9B` LM Studio, NVIDIA GPU (CUDA)	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), NVIDIA GPU (CUDA) set it yourself	`ACE-Step 1.5 XL turbo (music)` ACE-Step on Diffusers and PyTorch (AbstractMusic), NVIDIA GPU (CUDA) set it yourself	Read by the text model	Not available MLX-Gen image generation needs MLX, and MLX runs only on Apple Silicon Macs (macOS, arm64); set output.image to an image engine this host runs: diffusers (included with abstractcore[gpu]), sdcpp (stable-diffusion.cpp, included with abstractcore[apple] and abstractcore[gpu]) or a cloud image provider	Not available MLX-Gen video generation needs MLX, and MLX runs only on Apple Silicon Macs (macOS, arm64); no other local engine in AbstractFramework generates video (AbstractVision's Diffusers video path is disabled, stable-diffusion.cpp has none); the remaining option is an OpenAI-compatible video endpoint (AbstractVision OpenAI-compatible backend)
Linux or Windows, processor only	`Qwen3.5 9B` LM Studio, processor	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	Not available ACE-Step 1.5 XL turbo (music) needs about 10.9 GiB of memory while it generates (estimated), and this computer can give a model about 10.0 GiB; use a computer with more memory, or a cloud music backend (acemusic or elevenlabs-music, with its API key)	Read by the text model	Not available MLX-Gen image generation needs MLX, and MLX runs only on Apple Silicon Macs (macOS, arm64); set output.image to an image engine this host runs: diffusers (included with abstractcore[gpu]), sdcpp (stable-diffusion.cpp, included with abstractcore[apple] and abstractcore[gpu]) or a cloud image provider	Not available MLX-Gen video generation needs MLX, and MLX runs only on Apple Silicon Macs (macOS, arm64); no other local engine in AbstractFramework generates video (AbstractVision's Diffusers video path is disabled, stable-diffusion.cpp has none); the remaining option is an OpenAI-compatible video endpoint (AbstractVision OpenAI-compatible backend)
Intel Mac	`Qwen3.5 9B` Ollama, processor	`Supertonic 3 (TTS)` Supertonic on ONNX Runtime (AbstractVoice), processor	`Whisper base (speech to text)` faster-whisper on CTranslate2 (AbstractVoice), processor set it yourself	Not available ACE-Step music generation runs on PyTorch, and current PyTorch builds for macOS require Apple Silicon (arm64); the last Intel-Mac build is 2.2.2; set output.music to a cloud music backend (acemusic or elevenlabs-music, with its API key)	Read by the text model	Not available MLX-Gen image generation needs MLX, and MLX runs only on Apple Silicon Macs (macOS, arm64); set output.image to an image engine this host runs: diffusers (included with abstractcore[gpu]), sdcpp (stable-diffusion.cpp, included with abstractcore[apple] and abstractcore[gpu]) or a cloud image provider	Not available MLX-Gen video generation needs MLX, and MLX runs only on Apple Silicon Macs (macOS, arm64); no other local engine in AbstractFramework generates video (AbstractVision's Diffusers video path is disabled, stable-diffusion.cpp has none); the remaining option is an OpenAI-compatible video endpoint (AbstractVision OpenAI-compatible backend)

The same table, and the JSON it comes from, are in AbstractCore’s recommended models page. On your own machine: abstractcore models recommendations --host.

AbstractGateway · Web console

Everything about your gateway, in one place

Open http://127.0.0.1:8080/console. The console uses the address you opened it from, so it never asks for a gateway URL. The Technical details switch shows commands and ids throughout.

Models

Models that fit this machine

One card per model with every 4-bit and 8-bit build, its size, whether it is downloaded and whether it fits: Fits, Tight, Partial offload, Too large, or Needs GPU limit on Apple Silicon, with the exact sudo sysctl iogpu.wired_limit_mb=… command to copy. Search Hugging Face from the same view.

Engines

Local engines

Ollama, LM Studio, MLX, llama.cpp and others on the gateway host: installed or not, running or not; install, start and stop, each confirmed with the command it runs.

Providers

Cloud and endpoint providers

OpenAI, Anthropic, OpenRouter, Portkey, LM Studio, Ollama and custom OpenAI-compatible endpoints, with write-only keys.

Multimodal

Routes for every capability

Default models for text, voice, images, video and more. A route whose engine is missing says so with the install command; voice pickers say why a provider lists no voices.

Apps

Install and open the apps

Each app opens in a new tab at /apps/<app>/, already signed in. Code can also open in a terminal on the gateway machine; the desktop Assistant opens on the gateway machine, in its menu bar (macOS) or system tray (Windows, Linux).

Network

Who can reach the gateway

This computer only (the default), the local network, or the internet, with every address to copy and Advanced: reverse proxy.

Resources

Memory, models, start at login

Memory and GPU meters, resident models (warm up, lock, unload), prompt caches, and the Gateway card: pause, update, restart, desktop icon and a real Start at login switch.

Workflows

Workflows and defaults

Pick the default agent workflow for each app interface, turn streamed replies on by default, import, export and delete workflow bundles.

Users & Sandbox

Users, entities and a sandbox

User records and token rotation, the summoned-entity roster, and a sandbox for quick chats and media generation against the configured defaults. A docs assistant in the top bar answers from the gateway's own documentation.

The web console's Models tab searched for llama: Llama 3.3 70B builds marked Tight and Needs GPU limit, with the sudo sysctl command that raises the Mac GPU memory limit, and Llama 4 Scout builds marked Too large — Fit verdicts on a 128 GB Mac: Tight, Needs GPU limit (with the command to run) and Too large.

The web console's network choice: Localhost only selected, with Local network and Internet options and the list of addresses a client can use — Who can reach the gateway: this computer, the local network or the internet.

The web console's Apps tab: Node.js ready, then Observer, Continuum, Code, Entity, Flow Editor and Assistant cards with Open buttons — The Apps tab: each app opens at `/apps/<app>/`, signed in.

AbstractGateway · Terminal console

The same console, over SSH

abstractgateway-console does what the web console does, with the same confirmations. It is a Rust application rendered by AbstractTUI, and it reuses the Models and Engines screens of AbstractCore’s terminal console. On a headless server it runs the setup guide, and it never tries to open a browser on a machine without a display.

Setup guide and thirteen screens

For an admin whose first run is not finished, it opens the guide: Connection → Setup → Engines → Providers → Routes → Models → Apps → Review, ending with Finish and a Start at login switch. Otherwise it opens in browse mode.

1–9, 0, A, N: Connection, Providers, Routes, Users & Entities, Runtimes, Workflows, Review & Test, Resources, Models, Engines, Apps, Network
N Network shows the saved mode next to what is running now, and offers the restart that applies it
A Apps over SSH shows the one-time link and the ssh -L forward instead of starting a browser
F2 docs assistant, F3 gateway host panel (pause, restart, update, start at login)
Finds the gateway from --gateway-url, else ~/.abstractframework/gateway.json, else 127.0.0.1:8080, and follows it to a new port

# on the server (the installer prints this line, filled in)
abstractgateway-console --gateway-url http://127.0.0.1:8080 --token <admin token>

# choose the mode at launch
abstractgateway-console --wizard
abstractgateway-console --browse

# print the About text without opening the interface
abstractgateway-console --about --gateway-url http://127.0.0.1:8080

Keys: Tab focus, Enter activate, Ctrl+G the guide, r refresh, q quit. Admin-only actions are refused before anything is sent, with the reason. Needs Rust 1.87+ to build.

The terminal console abstractgateway-console on its Setup screen: numbered screen tabs, this computer's memory and graphics, data folder, sign-in mode and start at login, and the eight guide steps — The terminal console's Setup screen.

The terminal console's Network screen: Localhost only selected, Local network and Internet options, the address list and reverse-proxy settings — The Network screen (`N`): the saved mode, what is running, and every address.

AbstractGateway · Commands you get

The console's command-line twins

The one-line installer puts every terminal command in ~/.local/bin, next to abstractgateway: the terminal console, AbstractCode's terminal client, and the AbstractCore, AbstractVoice, AbstractMusic and AbstractVision commands. The console's Apps page installs and updates abstractcode in that same folder.

abstractcode login --token <admin token>     # once, then: abstractcode
abstractgateway apps tui-command code        # on the gateway's computer: a one-time signed-in line
abstractgateway-config claim-url             # a new one-time console sign-in link
abstractcore models catalog --fits           # the Models tab from the command line

Every command and its opt-out →

AbstractCore · Web and terminal consoles

AbstractCore has its own consoles

When you use AbstractCore on its own, without a gateway, abstractcore serve starts its OpenAI-compatible server on 127.0.0.1:8000 and prints a one-time link to its web console at /console: see what this machine can run, browse and download models with fit badges, delete weights, and install engines such as Ollama with one confirmation. Every action shows its command. abstractcore-console (crates.io, v0.4.1) is the terminal version, rendered by AbstractTUI: the same Models and Engines screens, plus your default models, providers, API keys and capability routes. The gateway’s terminal console reuses these Models and Engines screens.

pip install abstractcore && abstractcore serve   # http://127.0.0.1:8000/console#claim=…
cargo install abstractcore-console                # Rust 1.87+
abstractcore models catalog --fits                # the same catalog from the command line

Where to next

Keep going

InstallOne line on macOS, Linux and Windows, or the Mac package. Start as a userChat, voice, automations and the Assistant in your first hour. Start as an architectNetwork modes, headless servers, security and data. AbstractGatewayThe control plane behind the gateway consoles. AbstractCoreThe library behind the AbstractCore consoles.