Skip to main content
Support models are configured in the Support AI section and completely separated from Axosun: their profiles, their keys and no tools. Support AI talks to customers and has no access to infrastructure activities. You don’t need to link it to Axosun.

Model profiles

The profile describes the specific model of the provider: OpenAI, OpenAI-compatible API or Anthropic, model ID and API key. The key is stored in a secure storage and not shown back. For Anthropic, you can not override the API address: a fixed address protects the key from sending to a swap host. Each profile can be checked with a button on the card: the panel makes a trial call and shows the result (“Verified”, “Error”). The profile assigned to the role can not be deleted, first remove the appointment. Model calls go through a proxy if it is set: existing, new SOCKS5 or HTTP, or through the Axottle node.

rolely

The role is responsible for one piece of work. Each is assigned the main model and, if desired, a backup; the backup cannot match the main one, otherwise the main backup path will not remain. The role can be turned off entirely. The role of “Observer” stands out: assigning a model to it is a complaint detector switch, there is no separate flag. Unlike other roles, it reads every move of the client, including the operator’s dialogues and the AI mode turned off, and never responds to the client, only retrieves a complaint signal. Details on the page Mass incidents.
Roles “Routing”, “Controler” and “Analytician” can already be assigned, but in the conveyor they are not yet involved: the appointment of the model on them now does not affect anything.

Degradation without a model

The lack of a model doesn’t break support. Without Respondent, AI modes are unavailable, ticketing goes to operators. Without Visual Analysis, screenshots are marked unrecognized. Without Summarization, resumes are collected mechanically from correspondence. The transition to a backup model is not hidden: the counter of such calls is visible in use.

Limits and use

The “Daily Call Limits” card limits the number of calls to models per sliding day, separately for each role and globally. The money limit is not consciously maintained: the module does not know the pricing of providers, and a limit that will never work would give a false sense of protection. The 24 Hours Usage card shows calls, tokens and backup models for each role.