Chat shell

Who's serving
the answer.

When you open the chat model picker, every entry carries a small coloured badge that identifies the provider behind that model — Ollama on your cluster, Ollama Cloud, Eldric Inferenced, OpenAI, Anthropic, xAI, HuggingFace, and so on. The same badge appears under every assistant message. The point is simple: two models with the same name can come from different providers and behave differently. The badge tells you which one is currently serving.

The colour convention

Each provider gets its own colour.

The badge is a coloured rounded square with a short code or vendor mark inside. Different colours for different providers — so you can scan the picker and spot the route at a glance, without reading the small text underneath. Two entries called gpt-oss-120b served by different providers will carry different badges, and the assistant chip under their replies will match.

The badges, by provider

Provider key.

O Ollama
An Ollama runtime on one of your cluster's inference workers. The most common path for open-weight local models.
O Ollama Cloud
The hosted Ollama service. Same protocol as local Ollama, different colour — request leaves your cluster to ollama.com.
EI Eldric Inferenced
The native GGUF / xLSTM runtime that ships with the platform. No Ollama dependency, no external process. The path for the smallest deployments and air-gapped sites.
L Local — other
vLLM, llama.cpp, MLX, TGI or Triton on one of your inference workers. Covers any local runtime that isn't Ollama or Inferenced.
AI OpenAI
OpenAI's API. Request and response transit OpenAI infrastructure.
A Anthropic
Anthropic's API (Claude models). External.
X xAI / Grok
xAI's API (Grok models). External.
🤗 HuggingFace
HuggingFace Inference API or TGI endpoints. External.
G Groq
Groq's high-speed inference API. External.
? Unrecognised
A custom or fallback backend the platform doesn't have a specific badge for. Hover the badge to see the raw provider id.

Other providers — Together, Mistral, Fireworks, Cohere, Perplexity, Azure OpenAI, NVIDIA NIM and any custom endpoints — get their own colour-coded variants as the platform learns about them. If you wire up a provider the platform doesn't know yet, you'll see the ? badge with the actual id available on hover.

Where the badge shows

Picker, active model, every answer.

The badge sits next to the model name in the picker so you make an informed choice before you start. The "active model" pill at the bottom of the composer carries the same badge while you type. Below every assistant turn, a small chip shows the provider that actually served that one message — proof of who responded. If the cluster reroutes mid-conversation (worker drops out, backend rate-limits you, you switch model), you see the badge change. The system never silently swaps providers between turns.

Two models, same name

Telling them apart.

If two entries in the picker share a model name (a common case with open-weight models that several providers host), look at the badge to tell them apart. O gpt-oss-120b is your local cluster's Ollama runtime; O gpt-oss-120b is Ollama Cloud. They're the same weights, but the inference happens in different places, on different hardware, with different latency characteristics — and the responses can differ subtly because of sampling or version drift between hosts.

The badge is the simplest way to answer "where is this answer coming from?" without leaving the chat.