On this page
VGraple CRM's AI features work out of the box on a free, hosted Groq model, but you can connect your own OpenAI or Anthropic key if you specifically want GPT-4o Mini or Claude Haiku to answer instead. This article covers adding a key, what changes once you do, and a real limitation worth knowing: embeddings for semantic search need an OpenAI key specifically, no matter which provider answers chat.
Bring your own AI key: first 5 of 6 steps
- 1Get an API key from your chosen
- 2Go to Settings, then AI & Flows
- 3Save
- 4If you now have more than one
- 5Check individual AI Response nodes and the
Before you start
- Only Owner and Super Admin roles can view and edit these keys, since the settings page is restricted to roles with secret-viewing access.
- Groq's key is free (no credit card, up to 14,400 requests a day) and is enough to use every AI feature in the product without adding a paid key at all.
- Your own OpenAI or Anthropic key is billed directly by that provider at its own published rates; VGraple CRM adds no markup on top.
- If your knowledge base needs semantic vector search rather than keyword fallback search, you need an OpenAI key specifically for embeddings, even if your chat model is Anthropic or Groq.
Steps
- Get an API key from your chosen provider. Anthropic's console, OpenAI's platform dashboard, or Groq's console (free, no card) all issue keys directly; links to each are shown next to the matching field in the app.
- Go to Settings, then AI & Flows. Paste your key into the matching field (Anthropic API Key, OpenAI API Key, or Groq API Key). Each field is masked by default; click the eye icon to reveal what you typed before saving.
- Save. The key is stored encrypted and takes effect immediately for new AI calls; nothing needs to be redeployed or restarted.
- If you now have more than one key, set a Default AI Model. This dropdown only appears once at least one key exists, and it controls reply suggestions, conversation summaries, and the provider a brand-new Flow Builder AI node starts with.
- Check individual AI Response nodes and the chat widget, if relevant. Each one can be set to a specific provider independently of your org default; adding a new key does not retroactively change an existing node's chosen provider unless you edit it.
- Add an OpenAI key separately if you want semantic search, even if OpenAI is not your chat model of choice. This is the one case where a second key genuinely changes behavior beyond just "which model answers."
An organisation standardized on Anthropic for its brand of writing, for example, would add only the Anthropic key for chat, and could stop there. But if that same organisation's knowledge base is large enough that keyword search starts missing genuinely relevant chunks, adding an OpenAI key too, purely for embeddings, upgrades retrieval quality without touching which model actually writes the replies.
What you will see
Once a key is saved, it becomes selectable everywhere a provider picker exists: the Default AI Model dropdown, individual AI Response node settings, the chat widget's AI provider, and the AI debugger's test buttons. A provider with no key shows as greyed out and disabled in the debugger, and is simply absent from any dropdown that only lists configured options. Removing a key later does not break anything that was using it; every code path that calls an AI provider falls back automatically to any other configured key rather than failing outright.
The fallback order favors the cheapest usable option first: an AI Response node whose saved provider now has no key checks Groq before OpenAI, and OpenAI before Anthropic, so a missing key degrades to a free reply rather than a costly one by default. This only matters as a background safety net; day to day, whichever provider you actually pick for a node or widget is the one that answers.
Settings and options
| Setting or field | What it does | Default |
|---|---|---|
| Anthropic API Key | Powers Claude Haiku in Flow Builder, reply suggestions, and the AI debugger | Not set |
| OpenAI API Key | Powers GPT-4o Mini in the same places, and is the only provider that generates knowledge base embeddings | Not set |
| Groq API Key (Free) | Powers a hosted model (GPT-OSS 120B) with no cost or card required | Not set |
| Default AI Model | Which provider answers reply suggestions, summaries, and new AI nodes by default | First configured provider |
| Per-node / per-widget provider | Overrides the org default for one specific AI Response node or chat widget | Inherits org default at creation time |
| Provider fallback | Automatically switches to another configured key if the chosen provider's key is missing or fails | Always on, not configurable |
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| The Anthropic or OpenAI field will not save | The pasted key is malformed or was copied with extra whitespace | Re-copy the key directly from the provider's dashboard and paste it fresh |
| Knowledge base retrieval feels imprecise despite having an Anthropic key | Semantic search needs an OpenAI key specifically for embeddings; an Anthropic-only setup falls back to full-text and keyword search | Add an OpenAI key purely to power embeddings, even if Claude remains your chat model |
| A flow's AI Response node still uses a provider you removed the key for | The node's saved provider setting did not change on its own; it now falls back automatically to another configured key at runtime | This is expected and not a problem; edit the node directly if you want its saved provider label updated too |
| The Settings, AI & Flows page redirects you elsewhere | Your role does not have secret-viewing access | Ask an Owner or Super Admin to add the key, or to grant your role that access if appropriate |
| Costs on your OpenAI or Anthropic account are higher than expected | Every AI Response node, reply suggestion, summary, and embedding call using that provider bills against the same key | Review which flows and widgets are set to that provider; switching high-volume, low-complexity nodes to the free Groq key reduces paid usage |
Related reading
See setting up the AI chatbot for where these keys get used day to day, and knowledge base sources for the specific OpenAI-only embeddings dependency in more detail.