Reflex uses Jev for 4 judgment calls in a 20-step ticket
Open-source ticketing agent Reflex hands judgment calls like config changes and pre-push diffs to Jev for typed answers with confidence, running the LLM once at about ₹0.16 per ticket.
An access ticket came in. Here's the workflow that closed it, step by step, with what each step cost.20 steps. 4 judgment calls go to @typesafeai 's Jev, not an LLM. The LLM runs once. Whole ticket: ₹0.16. Jev's share: under 1 paisa. What ran: 1. Code pulls the repo + ticket (free) 2. Jev triages: config change, 100% conf, 326 ms 3. LLM edits one line of config on a branch (30s, ~15 paise) 4. Jev reads the diff before the push: safe 95% 5. Code pushes, comments on the ticket 6. Spawns a monitor fot git and Jira The pattern: → code first, for anything with an algorithm → Jev for judgment: typed answers with a confidence, so routing is a rule, not a prompt → an LLM only where something must be written → hard budgets per run Routing is data: push_gate.safe.noul >= 0.7 → push The monitor is a workflow too: a cron, one check, two questions, ~40 paise a day. It reports back; the parent knows an update from a fresh ticket. Caveat for India: not all data can leave the country. Reflex routes per step, so sensitive steps can stay on an in-country model. Reflex is open source and very new. It WILL have bugs. Issues and PRs welcome, small ones included. Built on the Pi coding agent, any model via @OpenRouter , voice via @SarvamAI . MIT. http://github.com/kaustav1996/reflex