Compare models
Send one prompt to up to three models at once and compare answers, speed and cost.
Models
Add a model here and it shows up in the chat model picker right away.
All models by type
Every model on this server, local, hosted and voice, with its type. Types are guessed from the name; change one with its menu. Other apps read this list at GET /v1/catalog?type=… with an API key.
| Model | Provider | Type | API id |
|---|
Voice models voice worker
Open-source text-to-speech and voice cloning, run on this server's CPU by the voice worker. Apps use them through /v1/voice/… (the Voice Studio at console.todaytim.com does). The first synthesis after a restart loads the model, which takes about a minute.
| Engine | Type | Languages | License | Status |
|---|
Local models Ollama
Runs on your server. Pick a model that fits in RAM: roughly 1 GB per billion parameters (quantized) plus some headroom. Browse names at ollama.com/library.
Browse the Ollama library (latest models, live from ollama.com)
Browse Hugging Face GGUF (abliterated, uncensored and every other GGUF model)
Abliterated models are community copies where the "refusal direction" was removed from the weights, so they decline far fewer requests. Uncensored ones were fine-tuned on unfiltered data. Both can lose some quality versus the original, and they come with no safety layer: you are responsible for how you use them. Check each model's card and license.
| Model | Size | Params | Quant. | Speed |
|---|
Hosted models API
Large models (GPT, Claude, Gemini, Llama 70B+…) through any OpenAI-compatible API. Add a provider once, then enable the models you want.
:free models usable with a free account key. Google AI Studio (Gemini), Groq, Mistral and Cerebras also offer free tiers with rate limits.Where to find models
- Ollama library · local models; copy a name like
qwen3:4binto Download - Hugging Face GGUF (trending) · download any of them as
hf.co/<user>/<repo>:Q4_K_M - OpenRouter free models · get a key at openrouter.ai/keys
- Google AI Studio key · Gemini models, free tier
- Groq key · very fast open models, free tier
- LMArena leaderboard · which models rank best right now
Monitoring
Which app used which model, how much and for how long, and what share of the server it took.
Model calls
| App | Model | Calls | Tokens in / out | Model time | Per call | Speed | Server CPU | Model RAM |
|---|
One row per app and model. An app is the API key it uses (Settings → API access); Console chat is this console. Model time is the time the model spent answering. Server CPU is the load of the whole server while those calls ran, other apps included; Model RAM is what the local model held in memory.
Server load
Share of the whole server (all apps): CPU across every core, memory in use, and the part of that memory held by loaded local models.
Latest calls
| When | App | Model | Tokens in / out | Time | Speed | Server CPU | Result |
|---|
Settings
API access for other apps, usage and costs, security and backups.
API access OpenAI-compatible
Other apps (n8n, scripts, editors, Open WebUI…) can use every model of this console through one OpenAI-style endpoint. Create a key per app so you can revoke it on its own.
| Name | Key | Created | Last used |
|---|
Examples
Usage and costs
| Model | Provider | From | Requests | Tokens in / out | Cost |
|---|
Local models cost nothing. OpenRouter reports the exact cost; for other providers, check their dashboard (not every API reports token counts while streaming).
Security
Two-factor login: after the password, the console asks for a 6-digit code from an authenticator app (Google Authenticator, Microsoft Authenticator, 1Password, Bitwarden…).
Backups
Every day the console saves providers, API keys, settings, conversations, prompts and usage to a zip in its data volume and keeps the last 14. Download one to keep a copy off the server. Each zip contains restore steps. Downloaded models are not included: they can be pulled again.