The co-op's central value prop is member choice across differently-valued providers. The current picker lists model IDs alphabetically with no framing, so new members can't see the values dimension at all.
Context
The co-op's models page already defines the framing: privacy (architectural vs policy) and green energy as independent axes, plus sovereignty for PublicAI.
Vision-capable models are marked in the docs; the picker could also surface a small vision badge.
The chat chrome itself is now themed to the co-op's house style (cream/green, Figtree) via the custom.css hook; the picker is the next surface.
Constraints
Open WebUI's license requires attribution; UI modifications should stay license-compatible (the current theme approach is; a fork later would need thought).
Model list comes from the gateway dynamically — grouping must not require manual updates when models are added (group by the provider/ prefix).
Acceptance idea
Model picker groups by provider prefix with a values label and one-line explainer
New models appear in the right group automatically
Works in both light and dark mode with the co-op theme
Ticket: chat UX — value-based model picker
Body:
## What's the idea?
Improve the model picker in chat.inference.coop so models are grouped by **provider values** rather than shown as flat `provider/model` strings:
- **Private (Tinfoil)** — architecturally private (TEE enclaves)
- **Green (GreenPT)** — 100% renewable energy
- **Public (PublicAI)** — publicly developed, sovereign open models
Each group could carry a one-line explainer (drawn from the docs' model page, https://git.inference.coop/co-op/docs/src/content/docs/models.md) so members see the co-op's core idea — choosing AI by what you value — every time they pick a model.
## Why?
The co-op's central value prop is **member choice across differently-valued providers**. The current picker lists model IDs alphabetically with no framing, so new members can't see the values dimension at all.
## Context
- The co-op's models page already defines the framing: privacy (architectural vs policy) and green energy as independent axes, plus sovereignty for PublicAI.
- Vision-capable models are marked in the docs; the picker could also surface a small vision badge.
- The chat chrome itself is now themed to the co-op's house style (cream/green, Figtree) via the custom.css hook; the picker is the next surface.
## Constraints
- Open WebUI's license requires attribution; UI modifications should stay license-compatible (the current theme approach is; a fork later would need thought).
- Model list comes from the gateway dynamically — grouping must not require manual updates when models are added (group by the `provider/` prefix).
## Acceptance idea
- Model picker groups by provider prefix with a values label and one-line explainer
- New models appear in the right group automatically
- Works in both light and dark mode with the co-op theme
The value-based picker framing shipped via config-only changes, no fork needed:
Friendly names + value-tier emoji set in the portal's /v1/models handler (what Open WebUI reads): 🔒 Private (Tinfoil/TEE) · 🌱 Green (GreenPT/renewable) · 📖 Public (PublicAI/sovereign). The id field is untouched, so actual requests, keys, and API calls are unaffected — emoji is display-only, verified against the Open WebUI source (the picker binds to model.id, not the label).
Value-tier clustering via Open WebUI's native ui.model_order_list (confirmed consumed by main.py), ordered Private → Green → Public so the tiers sit contiguous in the flat list.
Same emojis in the docs (models.md legend + every model row), so picker and docs tell one story.
What we deliberately did not do: fork the ModelSelector component for true section headers + one-line explainers (the issue's full acceptance idea). That requires patching compiled Svelte and maintaining a fork across upstream updates — with license considerations — so it's deferred for member sign-off rather than done unilaterally. The current change gets the "values visible in the picker" goal at ~zero maintenance cost.
**Implemented (options 1 + 2) — closing.**
The value-based picker framing shipped via config-only changes, no fork needed:
- **Friendly names + value-tier emoji** set in the portal's `/v1/models` handler (what Open WebUI reads): 🔒 Private (Tinfoil/TEE) · 🌱 Green (GreenPT/renewable) · 📖 Public (PublicAI/sovereign). The `id` field is untouched, so actual requests, keys, and API calls are unaffected — emoji is display-only, verified against the Open WebUI source (the picker binds to `model.id`, not the label).
- **Value-tier clustering** via Open WebUI's native `ui.model_order_list` (confirmed consumed by `main.py`), ordered Private → Green → Public so the tiers sit contiguous in the flat list.
- **Same emojis in the docs** (`models.md` legend + every model row), so picker and docs tell one story.
What we deliberately **did not** do: fork the `ModelSelector` component for true section headers + one-line explainers (the issue's full acceptance idea). That requires patching compiled Svelte and maintaining a fork across upstream updates — with license considerations — so it's deferred for member sign-off rather than done unilaterally. The current change gets the "values visible in the picker" goal at ~zero maintenance cost.
Commits: code/member-portal `3cdc250`, co-op/docs `888ac8c` (+ follow-up cleanup `a647284` removing phantom model entries).
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Ticket: chat UX — value-based model picker
Body:
What's the idea?
Improve the model picker in chat.inference.coop so models are grouped by provider values rather than shown as flat
provider/modelstrings:Each group could carry a one-line explainer (drawn from the docs' model page, https://git.inference.coop/co-op/docs/src/content/docs/models.md) so members see the co-op's core idea — choosing AI by what you value — every time they pick a model.
Why?
The co-op's central value prop is member choice across differently-valued providers. The current picker lists model IDs alphabetically with no framing, so new members can't see the values dimension at all.
Context
Constraints
provider/prefix).Acceptance idea
Implemented (options 1 + 2) — closing.
The value-based picker framing shipped via config-only changes, no fork needed:
/v1/modelshandler (what Open WebUI reads): 🔒 Private (Tinfoil/TEE) · 🌱 Green (GreenPT/renewable) · 📖 Public (PublicAI/sovereign). Theidfield is untouched, so actual requests, keys, and API calls are unaffected — emoji is display-only, verified against the Open WebUI source (the picker binds tomodel.id, not the label).ui.model_order_list(confirmed consumed bymain.py), ordered Private → Green → Public so the tiers sit contiguous in the flat list.models.mdlegend + every model row), so picker and docs tell one story.What we deliberately did not do: fork the
ModelSelectorcomponent for true section headers + one-line explainers (the issue's full acceptance idea). That requires patching compiled Svelte and maintaining a fork across upstream updates — with license considerations — so it's deferred for member sign-off rather than done unilaterally. The current change gets the "values visible in the picker" goal at ~zero maintenance cost.Commits: code/member-portal
3cdc250, co-op/docs888ac8c(+ follow-up cleanupa647284removing phantom model entries).