Google (Gemini)
Get a key
Section titled “Get a key”- aistudio.google.com/apikey.
- Create API key.
- Paste into Workforce0.
Free tier limits
Section titled “Free tier limits”Gemini has a production-grade free tier:
- 15 requests / minute
- 1,500 requests / day
- 1M input tokens / month free
This is enough for a hobbyist or small team deployment out of the box. No credit card required.
Set the key
Section titled “Set the key”GEMINI_API_KEY=AIza...Or: Settings → AI providers → Google (Gemini) → Paste key.
Models used by default
Section titled “Models used by default”| Role | Default model |
|---|---|
| Planner | gemini-2.0-flash-exp |
| Critique + revise | gemini-2.0-flash-exp |
| Specialists | gemini-2.0-flash |
| Voice live | gemini-2.0-flash-live-001 |
Override via MODEL_<ROLE>_GOOGLE.
Voice dial-in
Section titled “Voice dial-in”Gemini Live is the backend for voice dial-in. You need a Gemini API key with Live API access enabled — it’s enabled by default on new projects. See Voice dial-in.
Rate limits
Section titled “Rate limits”The free-tier limits (15 req/min, 1.5k req/day) apply per API key. Workforce0 handles them:
- Dispatches are throttled to stay under 15/min.
- 429 responses are backed off.
- Sustained failures degrade Gemini in the AI Council.
Cost rough numbers
Section titled “Cost rough numbers”| Model | Free tier | Paid input $/1M | Paid output $/1M |
|---|---|---|---|
| Gemini 2.0 Flash | yes | $0.075 | $0.30 |
| Gemini 2.0 Flash-Lite | yes | $0.019 | $0.075 |
| Gemini Live (per minute) | yes | $0.10 / minute | — |
Vertex AI
Section titled “Vertex AI”Vertex AI (Google Cloud) is supported as an alternative to the Generative Language API. Set:
GEMINI_USE_VERTEX=1GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.jsonVERTEX_PROJECT_ID=your-gcp-projectVERTEX_LOCATION=us-central1Useful if you’re already on GCP and want billing + audit on the same invoice.
What breaks
Section titled “What breaks””Quota exceeded — retry at…”
Section titled “”Quota exceeded — retry at…””Free-tier exhaustion. Add billing on the Google Cloud console to unlock paid tier, or wait.
Token-counting drift
Section titled “Token-counting drift”Google’s token counter occasionally disagrees with ours by <1%. We use theirs (via their count-tokens endpoint) for billing-accurate numbers. No action needed.
Safety filters
Section titled “Safety filters”Gemini’s default safety settings block “dangerous content” (weapons,
self-harm, etc). Legitimate briefs referencing such content (security
research, policy work) get filtered. Workforce0 sets the safety
thresholds to BLOCK_ONLY_HIGH by default — tune via
GEMINI_SAFETY_LEVEL.
Data processing
Section titled “Data processing”Current policy:
- Free tier: inputs and outputs MAY be used to improve Google products. If that’s unacceptable, upgrade to the paid tier.
- Paid tier: inputs / outputs are NOT used for training.
Read the live policy at ai.google.dev/gemini-api/terms.