kimiflare

About

Kimiflare addresses token-cost opacity by running inference on the user's own Cloudflare Workers AI account and routing traffic through Cloudflare AI Gateway, which returns authoritative per-turn costs, cache-hit ratios, and per-feature breakdowns instead of estimates. The agent ships 262k-context models, image understanding, MCP tool extension, LSP integration, local SQLite memory, and veto-capable hooks at five lifecycle points managed through a /hooks catalog. Modes cycle between plan (read-only research), edit (approval per mutation), and auto. Individual developers and small teams use it to keep coding-agent spend inside a Cloudflare bill they already control.