Keepance is a "bring your own key" (BYOK) tool. You sign up for an account directly with one or more AI providers, get an API key, and paste it into Keepance. This means:
You only need ONE provider to use Keepance. Most people start with Claude. You can add others later. Or skip the cloud providers entirely and run Ollama locally for free.
sk-ant-)Cost: Claude Sonnet (the default model) costs about $3 per million input tokens and $15 per million output tokens. A typical New Business Kickoff workflow run costs around $0.05-$0.20. Heavy daily use is usually $5-$15/month.
Free credits: Anthropic gives new accounts $5 in free credits. That's enough for ~50 workflow runs.
sk-)Cost: GPT-4o costs about $2.50 per million input tokens and $10 per million output tokens. Comparable to Claude Sonnet for typical workflows. GPT-4o-mini is much cheaper (~$0.15 / $0.60) and fine for less demanding tasks.
Note: An OpenAI API account is separate from a ChatGPT Plus subscription. If you pay $20/mo for ChatGPT Plus, that does NOT give you API access, you need to set up billing separately on platform.openai.com.
Cost: Gemini 2.0 Flash has a generous free tier (up to 15 requests/minute and 1500 requests/day at no charge). Beyond that, paid usage is roughly $0.075 per million input tokens, the cheapest of the three. Worth using as a "draft" model and switching to Claude or GPT for the final pass.
Ollama runs open-source models on your own machine. No API key, no provider account, no per-token charge. The tradeoff is hardware: you need a reasonably modern CPU and enough RAM (16 GB is comfortable for the smaller models, 32 GB+ if you want to run something like Llama 3.1 70B). On Apple Silicon it's genuinely fast.
ollama pull llama3.2 (or qwen2.5, mistral, etc.)http://localhost:11434Cost: Free. Your electric bill, basically.
Tradeoff: Local models are honestly really impressive in 2026, but they're still a notch below Claude Sonnet for long-form writing and complex instruction-following. Great for drafts, summaries, and anything you'd rather not send to a third party.
| If you want… | Use… |
|---|---|
| Best writing quality, long-form output | Claude (Sonnet or Opus) |
| Reliable structured output, JSON, code | GPT-4o |
| Cheapest, free tier for low-volume use | Gemini Flash |
| Multiple perspectives on the same prompt | All four (Keepance supports multi-model comparison) |
| Total privacy, offline use, no per-request cost | Ollama (local) |
All three providers let you set hard spending caps on your account. We strongly recommend setting one, usage-based AI billing can get expensive fast if something runs in a loop.
A reasonable starting limit is $10/month per provider. Adjust up if you find yourself hitting it often.
Plugins (the ones you install from Settings → Marketplace → Plugins) that ask for the ai:invoke permission use the same API key you already configured. There's one budget, not one per plugin. Every AI call a plugin makes is recorded in Keepance's audit log, tagged with the plugin's ID, so you can see exactly which plugin spent what.
Before a plugin installs, Keepance shows you a permission consent dialog with plain-language risk labels. If a plugin asks for ai:invoke, you'll see that listed. If you deny it, the plugin can still install but the AI features won't work. Read the plugin permissions doc for the full picture.
If you suspect your API key has been exposed (committed to a public git repo, shared in a screenshot, etc.):
Compromised keys can be used to run up charges on your account, so revoke immediately if there's any doubt.
This is a deliberate design choice. By having you bring your own key:
The downside is the small setup hurdle. Once it's done, it's done, you don't think about it again.