Bringing your own AI key

AI features in Roadmap OS use your API key — not ours. You pick the model, you pay your own bill, you own your data.

Why BYO key?

Most PM tools that ship "AI features" use their own model and bake the cost into your subscription. That has three problems:

1. You don't control the model. If they change models, your AI features change.
2. Your data goes through their pipeline. For privacy-sensitive teams, that's a hard sell.
3. You pay for AI even when you don't use it. Cost is bundled.

BYO API key fixes all three. You decide the model. Your data goes directly to the provider you trust. Costs match your usage.

Supported providers

• OpenAI — gpt-4o, gpt-4o-mini, o1 family
• Anthropic — Claude family (Sonnet, Opus, Haiku)
• Google — Gemini family (1.5 Pro, 2.0 Flash, 2.5 Pro)

You can configure multiple providers and select different models for different features.

Setup

Settings → AI → Provider

For each provider:

1. Get an API key from the provider's console:

• OpenAI: platform.openai.com/api-keys
• Anthropic: console.anthropic.com
• Google: aistudio.google.com/apikey

1. Paste the key into the corresponding field in Roadmap OS
2. Click Test connection — validates the key is working
3. Save

Keys are encrypted at rest in our database (AES-256) and never shown back to anyone (only the last 4 characters are visible after save).

Per-feature model selection

You can route different AI features to different models. Settings → AI → Per-Feature Model:

Feature	Default model	Why
Competitive analysis	Claude Opus 4.5	Strong reasoning
Decision dissection	Claude Opus 4.5	Long-context reasoning
Voice transcription	Gemini 2.5 Flash	Cheaper for audio
Brand guide extraction	GPT-4o	Strong document parsing
AI assist (general)	Claude Haiku 4.5	Fast and cheap

Switch any feature to a different model in the dropdown.

Cost transparency

Settings → AI → Usage

Shows your AI feature usage by:

• Provider
• Model
• Feature
• Day / month

We don't add a markup — your provider charges you directly. We just route the requests.

Typical monthly costs for active users:

• Light usage (1-2 AI features per week): $1-3/month
• Moderate (daily AI usage across features): $5-15/month
• Heavy (every initiative gets AI analysis): $20-50/month

Privacy notes

What we send to your AI provider:

• The specific text you ask the AI feature to process
• A system prompt describing the task

What we don't send:

• Your full database
• Other users' content
• Account credentials

What the AI provider does with it:

• OpenAI: by default, API calls are not used to train models (per their data policy)
• Anthropic: same — API calls aren't training data by default
• Google: same — Vertex AI / API calls aren't used for training by default

Confirm in your provider's settings. For especially sensitive data, use the "data residency" or "region" controls in your provider's account.

Disable AI features entirely

If you don't want any AI features:

Settings → AI → Disable AI features

Turns off all AI menu items, all AI-related buttons. The features remain disabled until re-enabled.

Troubleshooting

"Invalid API key" — re-paste the key. Common cause: copying with leading/trailing whitespace. "Rate limited" — you've hit your provider's rate limit. Either wait, or upgrade your provider plan, or switch the feature to a different model temporarily. "Model not available in your region" — some Anthropic models aren't available globally yet. Switch to a Google or OpenAI model.

Next steps

• Competitive analysis feature
• Decision dissection feature
• Voice transcription