agnts
Watch the demo
Pitch Day 3.0 · MVP live at agnts.uz

AI agents that take your tickets to a reviewed pull request.

agnts picks up a ticket from Jira or Plane, reads your code, asks what is unclear, codes and tests it in an isolated container, and opens the pull request only after you approve. Developers keep their time for architecture, scaling and business logic.

Watch the demo Open agnts.uz
May 2026Idea
June 2026First version, used on my own tickets since
20–25 mina day to keep projects moving while travelling
EN · RU · UZInterface languages

01Problem → Solution

AI writes the code. Developers still do everything around it.

Developers already code with AI, but every ticket still has to be read, explained, checked, tested and closed by hand. That routine takes most of the day.

The problem

Every ticket becomes a prompt

Read the ticket, dig through the code, then explain the task to the AI in enough detail that it gets it right.

Checking and testing by hand

Run the build, the linter and the tests, and fix what broke before anyone else sees it.

The paperwork

Push the branch, open the pull request, move the ticket to done.

The same mistakes again

The AI forgets your corrections, so you explain them again on the next ticket.

The solution: agnts

Reads the ticket and your code

Takes the ticket from Jira or Plane and asks short questions only where something is unclear.

Checks pass before you look

Codes in an isolated container, runs your build, lint and tests, and fixes failures itself.

From ticket to pull request

After you approve the diff, agnts pushes, opens the pull request and updates the ticket.

Learns from every mistake

Explain what went wrong once. agnts fixes it and keeps the lesson for the next tickets.

02Team

Built by an engineer who uses it on real work.

I build agnts on my own as a solo founder. A few programmer friends test it on their own work and in the mobile app.

Rustambek Abdusalomov

Founder · Engineer and team lead

Over 5 years in software engineering and team leading. Designed and built agnts end to end: server, admin panel, website, mobile app and infrastructure.

Skills

  • Backend architecture (Go)
  • Frontend (Vue, Nuxt)
  • Mobile (Flutter)
  • PostgreSQL
  • Docker and DevOps
  • AI agents (Claude Code, MCP)
  • API design (OpenAPI)
  • Team leading

Technologies used

ServerGo, Clean Architecture, REST API with Server-Sent Events, MCP server
DatabasePostgreSQL with migrations and row-level workspace isolation
AI agent runtimeClaude Code (Sonnet, Opus, Haiku) in one Docker container per run
Admin panelVue 3, Vite, TanStack Query
WebsiteNuxt 3, static build in English, Russian and Uzbek
Mobile appFlutter for iOS and Android, push notifications, Telegram bot
IntegrationsJira, Plane, GitHub, GitLab, Telegram
InfrastructureDocker Compose, Caddy (HTTPS), egress allowlist proxy, nightly backups, a Rust CLI installer

How agnts started

I wrote tickets over breakfast. By evening, they were ready for review.

  1. July 2026

    The trip to Turkey

    In July 2026 I went to Turkey on a trip. Every morning I spent just 5–10 minutes writing a few tickets, and when I came back to the hotel in the evening, they were waiting for me as tested pull requests. Reviewing them took only 10–15 minutes. So 20–25 minutes a day was enough to keep my project work moving, and I can honestly say it made a big difference to how good the trip was. That week I was truly thankful to myself for building agnts.

  2. May – June 2026

    How it started

    I started it in May, after seeing how much of a developer’s day still goes to reading tickets, explaining them to AI, then checking and testing the result. The first version started working on my own tickets at the end of June, and it has been working on them ever since.

  3. After the trip

    Friends and the startup

    When I came back, I told my programmer friends about it, and they got interested. They wanted to try it too, so I opened accounts for them, and soon after, every one of them gave me honest, positive feedback. That’s when agnts became a startup: I concluded that other developers should also spend their time on architecture, scaling and business logic rather than on repetitive daily tasks. I saw for myself how much it raises productivity, and how it lets projects be built with better, more polished solutions.

  4. Now

    The mobile app

    The mobile app is now in testing. I’m building it so the whole process can be managed from a phone. Once testing is done, I plan to launch it in full so everyone can use it.

Rustambek, founder of agnts

03Why me

Why I can solve it.

The problem comes from my own daily work, and the product already runs on real tickets.

The builder is the user

I have used agnts on my own tickets since June 2026. On a trip to Turkey I spent 20–25 minutes a day: writing tickets in the morning and reviewing finished pull requests in the evening.

Years inside the ticket workflow

Over 5 years of engineering and leading teams: I know where a developer's day goes, from vague tickets to review rounds.

Already working, not a mockup

The product is live at agnts.uz with a free demo. Programmer friends use it on their own work and gave very good feedback.

Built for safety and the local market

Nothing ships without human approval, every run is isolated, and the interface and support are ready for Uzbekistan, in Uzbek and Russian.

04Roadmap

Idea → Prototype → MVP → Launched.

agnts is at the MVP stage: live, free to try, and in use by its first developers.

  1. Idea

    Done
    May 2026
    • Automate the path from ticket to pull request with AI agents
    • Architecture and implementation plan

  2. Prototype

    Done
    June 2026
    • Jira ticket to a tested pull request
    • Isolated runs, checks and approval
    • Used on my own tickets since
    • On a trip, projects kept moving with 20–25 minutes a day

  3. MVP

    Current stage
    Now
    • Live at agnts.uz, free demo
    • Jira, Plane, GitHub, GitLab
    • AI review, previews, learning
    • Programmer friends using it on their work
    • Mobile app in testing with friends

05Implementation

How I build and run it.

Every ticket goes through the same pipeline. AI does the work in an isolated container; a person decides at every point that matters.

  1. Ticket in Jira or Plane
  2. Analyse code
  3. Answer questions
  4. Code in container
  5. Build, lint, tests
  6. AI review and preview
  7. Approve
  8. Pull request
  9. Learn
AI agentDeveloper decidesagnts server

AI tools and solutions

SonnetCoding

Writes the change. Failing build, lint or tests go back to the agent, up to twice.

OpusHard cases

Hard retries and reviews of complex or large changes.

HaikuHelpers

Summaries, translations and small helper steps.

Claude Code as the agent runtime

Each run gets a fresh Docker container with the repository and its toolchains. It can reach only Claude, the git host and package registries.

Asks before it guesses

An analysis run reads the ticket and the code first and turns anything unclear into short questions with options.

Self-review and red flags

A separate review run reads the diff, and a scan flags skipped tests, removed assertions and silenced linters before a person looks.

Learning from feedback

After a push or a change request, agnts distils lessons and suggests team rules. A person approves them before they reach any prompt.

Visual previews

For front-end work, a preview run opens the changed pages and takes screenshots at desktop and phone width.

Repository brief and replays

A brief of each codebase gives the agent context; replaying done tickets measures each new prompt version.

Stages

  1. Core agent loop

    Jira adapter, orchestrator state machine, container runner, checks and retries, GitHub pull requests.

  2. Daily-driver admin panel

    Live run logs, diff review, approvals, plan-limit tracking, workspaces, three languages.

  3. Integrations and agent intelligence

    Plane, GitLab, Telegram, MCP server for Claude chat, AI review, previews, learning from feedback.

  4. MVP with first users (now)

    Free demo at agnts.uz, friends using it on their work, mobile app in testing.

  5. Launch

    Paid plans, mobile app in the app stores, new connectors from user requests, and more AI models besides Claude, such as ChatGPT and Kimi.

06API access

Everything in agnts is an API.

The admin panel, the mobile app and Claude chat are all clients of the same API, so teams can script agnts too.

REST API

Described by an OpenAPI 3 contract, with bearer tokens and live updates over Server-Sent Events.

MCP server

Connect agnts to Claude chat, Claude Code or Claude Desktop and manage tickets in conversation, with a confirmation step before every change.

# A few operations from the OpenAPI contract
GET  /tickets                        # tickets and their state
GET  /tickets/{key}/events           # the run's live log
POST /tickets/{key}/answers          # answer agnts's questions
POST /tickets/{key}/approve          # approve: push and open the pull request
POST /tickets/{key}/request-changes  # send it back with feedback

See agnts work on a real ticket.

The demo shows a ticket going from Jira to a reviewed pull request, and the working product you can try today.

Go to the demo