AgentPlayground never hardcodes a single provider. You can mix and match — the coordinator on
Claude, one team on a free NVIDIA model, another on local Ollama — all in the same install, and
switch any of them at any time.

| Provider | Cost | Where your key/model comes from |
|---|---|---|
| Anthropic (Claude) | Paid | console.anthropic.com |
| OpenAI | Paid | platform.openai.com |
| NVIDIA | Free | build.nvidia.com |
| Ollama | Free, local | runs on your machine, no key |

Every model picker in the app — the main chat header, a playground's chat, a meeting participant's
model, an agent's pinned model — has the same shape: pick a provider tab, pick a curated model
from the shortlist, or type any model id that provider actually serves into the custom-model
field at the bottom and press Enter. The curated lists in lib/model-catalog.ts are a shortlist
for convenience, not a restriction — nothing in the app rejects an unrecognized model id at
picker-time, it just tries the request.

Two places, same fields, same behavior everywhere:

• First-run setup wizard — validates each key with a cheap request before letting you continue.
• Settings → API Keys — add, replace, or check the status of any key at any time. Each field shows whether it's set and whether it came from your .env file or was saved here (via Settings takes priority for that key).

Keys are encrypted (AES-256-GCM) before they're written to the database and only used for requests
your own install makes — they never leave your server or get sent anywhere but the provider you
configured.

An individual agent on a team can be pinned to a required model (for example, "this agent must
always use Opus") independent of whatever the chat window's picker is set to — see the **Set an
agent's model** entry in Prompt Templates, or drive it directly from
Team settings.

If nothing is configured, the app tells you exactly what to do next instead of failing silently:
"No AI provider available. Add an API key in Settings → API Keys (Anthropic, or NVIDIA's free key
from build.nvidia.com), or start Ollama for local models." The free-tier fallback chain (Anthropic
→ NVIDIA → OpenAI → Ollama) means a genuinely free install — NVIDIA key only, or Ollama only — is
fully supported, not a second-class path.