AI agent
Control Agent usage and cost
Use provider-side limits, Kition Cloud credits, smaller context, explicit model choices, and the Stop control without relying on nonexistent budget-routing features.
Updated
By Kition Docs
Verified against Kition 0.1.41 (d3e1b931) on 2026-08-29
Know who meters the request
Do not use a fixed “cost per task” estimate from old documentation. Check the selected provider’s current usage records and Kition credit balance.
- Kition Cloud usage consumes Kition credits shown in the account and billing surfaces
- Bring-your-own-provider usage is billed and rate-limited by that provider or gateway
- Tool loops can make several model turns for one user request
- Model pricing, token rules, and quotas can change without a Kition release
Controls that exist
- Choose the active provider and model from the Agent model picker
- Attach only the documents, folders, and files needed for the task
- Use a new chat when old conversation history is no longer relevant
- Stop a task that is looping, using the wrong context, or heading toward an unwanted action
- Set spend limits and alerts in the external provider console when available
- Use Kition Cloud credit balance and top-up controls for hosted models
Controls that do not exist in 0.1.41
- No Agent monthly dollar cap setting
- No per-session or per-tool token budget control in the Desktop UI
- No JSON model-routing rules by tool, task, context length, or subagent
- No Pro-only automatic routing between several external providers
- No Agent concurrency slider
Reduce unnecessary context
A focused context tray is easier to verify and can reduce repeated input. Prefer a few directly relevant documents over a broad folder when the task is narrow.
Local analysis sources are read-only and capped at eight, but their contents are still model-visible for the active run. Remove them from the tray when the task no longer needs them.
Use small checkpoints
- Ask for a read-only inventory before requesting writes
- Review a table plan before applying a large batch
- Split research, drafting, and final writes into separate turns
- Verify provider usage after the first representative task
- Stop and revise early instead of letting an incorrect tool loop continue
Implementation sources
Related pages
Switch the active provider and model
Understand the device-wide model selection that Kition 0.1.41 saves when you sync a provider or use the Agent model picker.
Agent chat basics
Open a workspace-scoped chat, attach only the context you need, choose a model, and review the streamed work.
Agent tool calls
Read the actual runtime tool names, availability signals, inputs, outputs, errors, and changed-file results shown in the Agent timeline.