Autocloz + OpenAI
OpenAI's GPT models can run every AI feature in Autocloz when you add your OpenAI API key in Settings → AI & Models and set OpenAI as the active provider. Autocloz then uses GPT for email drafts, per-lead personalization, reply classification, reply drafts, lead extraction, ICP-fit enrichment and AI lead lists. The offline fallback list starts at gpt-4o and gpt-4o-mini; with a valid key, Autocloz lists the models your account can actually use.
What you can do with OpenAI
OpenAI plugs into the AI replies & summaries stack — bring your own api key, scoped to your workspace.
Drafting and personalization
Generate a cold-email subject and body from campaign context, or rewrite one email for one lead. You review before it is used.
Enrichment and ICP fit
Prospecting enrichment asks the model for seniority, department and an ICP-fit score, which helps you order who to contact first.
Channel-aware rewrites
Channel personalization adapts a message to SMS, WhatsApp, LinkedIn or call-script length and tone, and can be switched per channel.
How to connect OpenAI
Bring your own API key — you keep control of the account, the spend and the data boundary.
- 1
Create an OpenAI API key
Create a project key in the OpenAI dashboard. Usage from Autocloz bills to that project.
- 2
Add it in Settings → AI & Models
Select OpenAI, paste the key and choose a model. Autocloz shows only the masked key afterwards.
- 3
Choose features
Switch off any feature you do not want, for example chatbot answers if you have not written a knowledge base yet.
- 4
Check usage
Compare the usage rows in Autocloz with your OpenAI usage page after a day of normal work.
Controls that apply before any GPT call
Every AI request passes the same gate in order: the workspace's master AI switch, the feature's own switch, the channel's AI switch where the feature is channel-specific, then either a bring-your-own-key check or a credit balance check. If any of these fails, no request is sent to OpenAI and you get a specific error such as ai.feature_disabled or byok.no_key. That order is why switching off chatbot answers, for example, stops that cost completely without affecting reply drafts.
- Master switch off: nothing runs.
- Feature off: that feature returns ai.feature_disabled.
- BYOK with no key: byok.no_key.
- Credits mode without balance: a recharge prompt.
What "lead scoring" with GPT means here
Autocloz has two different scores and it is worth keeping them apart. The CRM's lead score is rule-based and moves with activity such as opens, replies and deal-room views; see lead scoring. The AI piece is prospecting enrichment, where the model reads a lead's title and company and returns seniority, department and an ICP-fit score. That runs on whichever provider is active, so OpenAI is one option, not a requirement.
What OpenAI does not do in Autocloz
The earlier blurb for this page said GPT powers voice-agent "brains". That is not true today. The voice-agent editor lists an OpenAI Voice option, but calls from those agents are not wired to OpenAI. The only AI voice path that places real conversational calls is Sarvam over a Telnyx sender, described on the AI agents page.
AI replies are drafts. Autocloz does not auto-send what GPT writes, which keeps a person in the loop for every message that reaches a prospect. The reply workflow is described on AI replies.
What OpenAI does not do in Autocloz
- No AI voice calls on OpenAI.
- No automatic sending of AI-written replies.
- No fine-tuning or custom model upload through Autocloz; you pick from models your key can list.
Troubleshooting OpenAI
- Requests fail right after adding the key.
- New OpenAI projects sometimes have no billing set up, so every call is refused. Add billing in OpenAI and retry.
- Credits mode asks you to recharge.
- Your workspace is in credits mode, not bring-your-own-key. Either top up credits or ask an admin to switch mode.
Part of the AI replies & summaries stack
OpenAI is one of the AI providers you can bring to power replies, summaries and assist features. They plug into the same AI surfaces — you choose which key drives them.
Explore the AI replies & summaries featureRelated integrations
AI reply drafts and summaries with your own Anthropic key.
View integrationLow-latency LPU inference for fast AI replies.
View integrationVoice for AI agents: queued, not live. Live AI voice runs on Sarvam over Telnyx.
View integrationFrequently asked
Which GPT model should I pick?
gpt-4o-mini is enough for classification and short drafts. Use a larger model if draft quality matters more than cost.
Is my OpenAI key shared with other workspaces?
No. It is stored encrypted for your workspace and used only for its requests.
Does OpenAI train on my Autocloz data?
That is governed by OpenAI's API data policy for your account, not by Autocloz. Autocloz sends only what each feature needs and does not opt your data into anything.
Run OpenAI inside Autocloz
Bring your own account, plug it into the AI replies & summaries channel, and keep a free-forever CRM for 5 users, unlimited on paid.