How Do I Use My Own LLM API Key in Gumloop?
Last updated: July 31, 2026
Connect your own LLM provider API key (OpenAI, Anthropic, Google Gemini, Perplexity, xAI, DeepSeek, or Fireworks AI) and Gumloop charges 50% fewer credits for AI model usage in both agents and workflow AI nodes. BYOK requires a Pro plan or higher — add your key on the Connectors page, share one with a team, or set an organization-wide key on Enterprise.
Before You Start
- You need an API key from your provider (e.g. OpenAI, Anthropic, Google AI Studio, xAI, or Fireworks AI).
- BYOK requires a Pro plan or higher.
- Organization-wide keys are an Enterprise feature and can only be set by an organization admin.
Where to Add Your API Key
You can add an LLM key at three levels. Which ones are available depends on your plan.
| Level | Who It Applies To | Plan | Where to Add It |
|---|---|---|---|
| Personal | Only you — your own runs, in your personal space and in any team you belong to | Pro or higher | Settings → Connectors |
| Team | Everyone in that team, without each person managing their own key | Pro or higher | Settings → Teams → [your team] → Connectors |
| Organization | Every member of the organization | Enterprise | Settings → Organization → API Keys & Proxies |
How priority works: Gumloop checks for an organization key first, then a team key, then your personal key. The first one it finds is used.
Because of that order, an organization key overrides both team and personal keys for that provider — which is how Enterprise admins enforce a single, centrally billed key.
Steps to Add Your Key
- Go to the page for the level you want (see the table above).
- Click Add Connector. On the organization API Keys & Proxies page, click Add Key instead.
- Pick your provider (OpenAI, Anthropic, Google Gemini, Perplexity, xAI, DeepSeek, Fireworks AI, or Azure OpenAI).
- Paste the API key and save.
Gumloop starts routing matching model calls through your key on the next run — there is nothing to enable on the agent or workflow itself.
If you want the whole team on one key instead of everyone adding their own, see How Do I Add Shared Team Credentials (Team Apps)?.
How BYOK Affects Credit Costs
AI usage in Gumloop is billed by token usage, not by fixed per-call tiers. BYOK cuts that AI portion in half — you pay your provider directly for the tokens, and Gumloop charges half the credits.
| What You're Running | Without BYOK | With BYOK |
|---|---|---|
| Agent chats (the Chat & Reasoning portion) | Token-based cost | 50% fewer credits |
| Workflow AI nodes (Ask AI, Analyze Image, Generate Report, etc.) | Token-based cost | 50% fewer credits |
| Image generation | 30 credits per image | 15 credits per image |
| Audio transcription | Roughly 1–2 credits per minute | Half the per-minute rate |
What's not affected: tool call credits, the fixed per-node costs of non-AI workflow nodes, and the 1-credit base cost of a workflow run. The discount applies to AI model usage only.
For the full cost breakdown, see Credits in the docs.
Which Models Your Key Covers
A key only discounts the models that provider actually serves. An OpenAI key does nothing for a Claude run, and vice versa — so if you switch your agent's model, check that you have a key for the new provider.
- OpenAI — GPT models
- Anthropic — Claude models
- Google Gemini — Gemini models
- xAI (shown as SpaceXAI in the model picker) — Grok models
- Perplexity — Perplexity models
- DeepSeek — DeepSeek models called directly
- Fireworks AI — the open models Gumloop serves through Fireworks: DeepSeek V4 Flash, DeepSeek V4 Pro, Kimi K3, Kimi K2.7 Code, GLM-5.2, MiniMax M3, and Qwen3.7 Plus
- Azure OpenAI — if you run OpenAI models through your own Azure deployment
Open models that Gumloop does not serve through Fireworks — GPT-OSS 120B, Qwen3.5 397B, and Kimi K2.6 — are not discounted by a Fireworks key.
Not sure which model your agent is on? See How to Choose the Right AI Model for Your Agent.
How Do I Know It's Working?
- Run your agent or workflow once after saving the key.
- Open Chat Details from the chat header (or the run in Settings → Usage & Limits) and compare the AI portion of the cost to a comparable earlier run.
- The AI model charge should be about half what it was. Tool call credits stay the same.
If the cost hasn't changed, the most common causes are a key for a different provider than the model you're running, or an organization key that is taking priority over yours.
Security
- Encrypted at rest — keys are encrypted before storage.
- Auto-redacted — key values never appear in logs, run outputs, or agent responses.
- Scoped access — personal keys stay private to your account; team and organization keys are shared only with that group.
Still Need Help?
If this didn't resolve your issue, reach out to support at support@gumloop.com.