Does anybody else feel Now Assist is plain terrible?

jvasquezvil
Kilo Contributor

‎09-24-2026 04:45 PM

Does anybody else feel Now Assist is plain terrible?

Is a terrible mix of plugins, is convoluted, is way too basic and it has hidden flags to return cut down token generations as much as possible?

Like in "Extract information from documents use cases"

 

The call to the LLM engine has a clear "max_completion_tokens" to cut down compute and return a watered-down version of what you need.


  • 933 Views
3 REPLIES 3

abirakundu23
Giga Sage

‎09-24-2026 09:37 PM - edited ‎09-24-2026 09:37 PM

Hi @jvasquezvil ,

Follow documentation for Tokens : https://www.servicenow.com/docs/r/intelligent-experiences/now-assist-skill-kit/configure-skill-promp...

Please check Now Assist skill configuration, LLM/model, and token limits. The max_completion_tokens value may be a skill/model.

If the value is hard-coded in an OOB implementation, it would be worth raising a ServiceNow HI case to confirm whether this is expected behavior or a product limitation.

Please mark helpful & correct answer if it's worthy for you.

 

0 Helpfuls

Suggy
Giga Sage

3 weeks ago

Yes its a terrible terrible terrible...... ultra terrible product.

0 Helpfuls

sanjaygupta
Tera Guru

2 weeks ago

Hi Mark,

This limit is documented rather than hidden, although the documentation does not explain the rationale behind the default values.

In Now Assist Skill Kit, the prompt settings include Maximum response tokens and Maximum request tokens, and the documentation states that these limits are set by the admin for controlled usage. The Australia documentation adds that when the Now LLM Service is used, the maximum response tokens value is 1000. Switching to another provider may change this. The Now LLM model card also notes that outputs are constrained by token length and long outputs may be truncated.

For document extraction, Document Intelligence has its own limits. For example, image mode supports roughly 10 to 50 pages per file, which is not a hard limit. Dense text or large images consume more tokens, and the documentation recommends increasing the page count gradually to find the effective limit. A Community reply on a similar thread also notes that larger files can hit limits in Document Intelligence or the Now LLM.

To review your configuration:

  1. Open All > Now Assist Skill Kit > Home, select the skill, and go to Configurations or Manage prompt > Prompt settings.
  2. Check the Model and the Maximum response tokens value.
  3. If you are on the Now LLM Service, try a different supported model.
  4. Review the Document Intelligence use case and check whether image mode or table extraction is enabled, as each has its own page limits.
  5. Test with smaller files and raise the page count step by step.
  6. Review sys_generative_ai_config. A Community post indicates response max tokens can be set there for the prompt in use, while the same thread notes request tokens cannot be changed. This is a community tip rather than documented behavior, so test it in a sub-production instance first.

Note that tokens consume compute and, depending on your contract, assist usage, so raising limits may increase cost and latency.

The model that you are using determines whether the 1000-token cap applies.

Sources:

0 Helpfuls