FAQ – AskAI Failures Caused by Token Limits, Context Size, or Model Overload

This error only happens on databases using custom AI API keys (OpenAI or Gemini's). Custom API keys are currently required in Odoo SH and On-Premise. SaaS databases do not currently require users to input a custom AI API key since Odoo offers a free tier when this option is not configured. However, if SaaS users input their own API keys, this issue may come up.

General guidance: 
- If you experience similar behaviour in Odoo SaaS AND you do not have any custom API keys in use, this is not normal — please report it to the support team.

• As a reminder, Odoo AI is only available in the Enterprise version (paid), not in the Community version.
  
  

1. What are the common symptoms?

Users may report one or more of the following messages:

• “Sorry, I couldn’t process your request right now. Please try again later.”

• Invalid Operation – “The model is overloaded. Please try again later.”

• Errors appear regardless of model (GPT / Gemini, Lite or Pro).

• Simple queries (CRM, invoices, leads, etc.) intermittently fail.

These symptoms are typical when the AI provider rejects the request due to token/context limits or throughput restrictions.

2. Why does AskAI fail even with simple questions?

Even simple questions can generate large internal prompts when the agent has:

• Assigned topics

• Multiple tools enabled

• Database-heavy queries

AskAI sends a combined payload containing:

• User prompt

• Agent system prompt

• Tool prompts

• Extracted database data

• Model-specific tokenization

This can result in very high token counts, exceeding the customer’s API tier.

3. Why does it work sometimes but not always?

Because token usage varies per request.

If the request is near the API token limit:

• Some calls stay below the cap → they succeed

• Others exceed the limit → they fail

This produces intermittent failures, identical to what Synkiria observed.

4. How do we confirm the issue?

Check the Odoo server logs.

Each AI call logs the token count.

Example:

• Tokens sent: 31,690

• API limit: 30,000 (OpenAI Tier 1)

• → The provider rejects the request

• → AskAI returns a generic error

Any log showing tokens above the provider’s limit confirms the cause.

5. Why does Odoo show a generic “model overloaded” message?

If the AI provider rejects the request before processing it, Odoo receives no detailed error code.

The fallback message is shown automatically:

• “Model is overloaded”

• “Cannot process now”

This is normal when hitting token caps, rate limits, or context overflows.

6. What can be done to fix it?

✔ Upgrade the API tier

Example for OpenAI:

• GPT-4o Tier 1 → ~30k tokens/s

  Often insufficient for multi-topic AskAI agents.

(Always verify current tier plans directly with your provider as they may change.)

✔ Use a more efficient model

Recommended: Gemini Flash 2.5

• Lower token usage

• Faster

• Better quality

• More cost-efficient

• Requires a Google Gemini API key

✔ Reduce complexity of agent topics (optional mitigation)

Fewer topics → fewer tools → fewer tokens.

7. Quick Support Checklist

Before contacting Odoo Support, please confirm:

• ✔ You are using a custom API key

• ✔ Agent has topics

• ✔ You are on a low API tier

• ✔ Recommendation: switch to Gemini Flash 2.5 or upgrade the API tier

If all conditions match → This is a token-limit case.