Choose the AI models Dopia uses
Owners and admins choose which AI model writes deal memory, drafts emails and runs routines in Dopia, set its effort level, and see how that affects credits.
Dopia lets workspace owners and admins choose which AI model does each kind of unattended work: writing deal memory, drafting the emails in your queue and running routines. The setting lives in Organization settings → AI models. Every job starts on a platform default, Claude Haiku 5.5 today, and any change applies from the next run.
Before you begin
You need the Workspace owner or Admin role. Members can't open Organization settings, so the AI models page isn't available to them.
Change the model for a job
Open AI models
Click your name at the bottom left of the sidebar, choose Organization settings, then AI models.
Find the job
Under Models, each job has its own row: Deal memory, Email drafts and Routines.
Pick a model
Open the job's model menu and choose a model. Platform default follows Dopia's choice for that job, including if Dopia changes it later.
Set the effort
Next to the model, open Effort and pick a level. The menu lists only the levels that model supports and marks its Default.
Dopia confirms with "AI settings updated — applies from the next run." Picking a different model puts the effort back on that model's default.
What each job does
| Job | What it does | Platform default |
|---|---|---|
| Deal memory | Writes and updates each deal's memory, a judgment the whole team reads | Haiku 5.5, effort Extra |
| Email drafts | Drafts the emails in your queue, text customers read word for word | Haiku 5.5, effort Medium |
| Routines | Runs every routine that doesn't pin a model of its own | Haiku 5.5, effort High |
What deal memory tracks and Automate recurring work with routines explain the work these jobs do.
Which AI models can I choose in Dopia?
| Model | Provider | Effort levels |
|---|---|---|
| Sonnet 5.5 | Claude | Low, Medium, High, Extra, Max |
| Opus 5.5 | Claude | Low, Medium, High, Extra, Max |
| Haiku 5.5 | Claude | Low, Medium, High, Extra, Max |
| GPT-5.6 Terra | OpenAI | Low, Medium, High, Extra, Max |
| GPT-5.6 Luna | OpenAI | Low, Medium, High, Extra, Max |
| DeepSeek V4 Pro | DeepSeek | Low, High, Max |
| DeepSeek V4 Flash | DeepSeek | Low, High, Max |
| GLM-5.3 Flash | Z.ai | Low, High, Max |
| GLM-5.3 | Z.ai | Low, High, Max |
The list in the model menu is always the current one; Dopia adds and retires models over time.
How the model and effort affect credits
Dopia credits follow the cost of the work. A larger model, or a higher effort level, does more thorough work but takes longer and uses more credits for the same job. In the Effort menu, Max is marked "4× or more usage" when it isn't the model's default.
After a change, check Organization settings → Usage and filter by Work type to see what each job now costs. Dopia plans, seats and credits explains how credits are counted.
Other settings on the AI models page
- No approval needed for record changes, under Approvals, is off by default. When your assistant's routines or your sales agents' automations want to replace values already in the CRM, they normally wait for approval. Turn this on to apply those changes straight away; each one is marked "Auto-approved" in the run's log. Changes to tables and fields, and possible duplicate records, still ask.
- AI output language, under Output language, sets the language Dopia writes shared documents in, such as deal memory: English or 简体中文. Everyone on the team reads the same copy. Your own chats follow the language in your Profile.
Models chosen somewhere else
This page sets the workspace's defaults for background work. Three other places pick a model for themselves:
- Your chats with your assistant use the model you pick in the message box.
- A routine can pin its own model on its page. Left on Workspace default, it follows the Routines row here.
- A sales agent's automations each have a Model setting on the agent's page, which follows Dopia's platform default unless you change it.
Common questions
- Who can change the AI model in Dopia?
- Only workspace owners and admins, in Organization settings under AI models. Members can't open Organization settings, so they can't change which models the workspace uses.
- Which AI model does Dopia use by default?
- Claude Haiku 5.5 runs all three jobs on this page by default, deal memory, email drafts and routines, each at its own effort level. A job left on Platform default follows Dopia's default if it changes later.
- Does changing the model redo work Dopia has already done?
- No. A new model or effort level applies from the next run. Deal memory, drafts and routine results written earlier stay as they are.
- Do larger AI models use more Dopia credits?
- Yes. Credits follow the cost of the work, so a larger model or a higher effort level uses more credits for the same job. Organization settings under Usage shows the effect by work type.
Up next
Manage sign-in and security
Sign in to Dopia with Google, Microsoft or an email code, see every device signed in to your account, sign sessions out, and learn how your data is protected.