PyModel

Our products.

These are systems we built and run ourselves, plus the developer tools we ship. Each one names the problem it solves and how far it has actually come. Some carry production traffic today. Others are still experiments, and the label says so.

View projects on GitHub
Flagship · always-on AI agentsIn development

Pythinker Bot

A team of digital coworkers that do real, multi-step work across applications, websites, and development environments.

For Teams that want to delegate whole jobs, not just prompts, and keep the work on their own machines.

Problem
Chatbots answer inside a conversation and stop when the tab closes. Real work spans tools, files, browsers, and days, and it needs somewhere persistent to happen.
What PyModel is building
An always-on AI agent product. Each Pythinker Bot runs inside its own persistent sandbox directly on your machine, with an isolated browser, filesystem, terminal, and development tools, available 24/7 without an active chat session.
What bots do
Research, software development, data processing, browser automation, testing, and recurring business workflows. Multiple bots coordinate and work in parallel, each in its own sandbox, on larger tasks.
Technical approach
Local-first execution instead of remote cloud computers: persistent state, local files and tools, and greater control over data. Bots surface decisions and request approval when human input is required, and learn reusable routines from instructions or demonstrations so repeated processes become persistent automations.
Backend systemsActive system

Operations backend

A Go backend for CRM and ERP workflows.

For Companies running sales, inventory, and billing in separate tools.

Problem
Companies run sales, inventory, and billing in separate tools, and the numbers stop agreeing with each other.
What PyModel built
A Go backend that puts customer records, orders, inventory, and invoicing behind one API.
Technical approach
PostgreSQL serves as the single source of truth across CRM and ERP workflows.
Agentic RAGActive system

Pixie Q

An agentic retrieval pipeline for grounded answers from a team’s own documents.

For Teams that need sources they can check instead of model guesses.

Problem
Teams need answers grounded in their own documents, with sources they can check, not model guesses.
What PyModel built
An agentic retrieval pipeline for questions over a team’s own documents.
Technical approach
Query planning, multi-step retrieval, grounded generation with citations, and answer verification.
Backend infrastructureProduction pattern

Commerce systems

A backend infrastructure pattern for checkout and merchant integrations.

For Teams whose checkout workflows depend on durable state, retries, and webhooks.

Problem
Checkout and merchant integrations fail when state, retries, and webhooks lack durable ownership.
What PyModel built
Go services backed by PostgreSQL state for commerce workflows.
Technical approach
Durable jobs, idempotency, and webhook handling keep integration boundaries explicit.
AI product conceptLabs

Pythinker Code

A coding-assistance concept for Python, TypeScript, React, and Go tasks.

For Technical teams that need a focused coding partner across languages.

Problem
Technical teams need a focused coding partner that works across languages without hiding uncertainty.
What PyModel built
A coding-assistance concept for Python, TypeScript, React, and Go tasks.
Technical approach
Precise generation, systematic debugging, optimization, and explanation without hiding uncertainty.
Medical & dentalLabs

Clinical practice systems

Software and AI integration for dental and medical practices, developed with direct clinical experience.

For Practices whose scheduling, records, imaging, billing, and patient communication live in separate tools.

Problem
Practices run scheduling, records, imaging, billing, and patient communication in separate systems, and AI assistance arrives without clinician review or clear patient-data boundaries.
What PyModel is building
An integration and AI-assistance pattern for clinical workflows: intake, documentation, recall and patient communication, and imaging-adjacent support that stays subordinate to the clinician. PyModel is led by a practicing dentist.
Technical approach
Practice-management integrations, clinician-in-the-loop review, audit trails, and explicit patient-data boundaries before any model output reaches a chart or a patient.
macOS menu bar app · freeIn development

ProxyPort

A menu bar app that shows every dev server on your Mac, and the coding agent that started it.

For Developers running coding agents and git worktrees who lose track of what is listening on localhost.

Problem
Coding agents and worktrees leave dev servers running everywhere: ports from deleted branches, servers slowly eating memory, and no record of which agent started which.
What PyModel is building
A native macOS menu bar app and a proxyport terminal UI. Each server shows its port, project, git branch, uptime, memory, CPU, traffic, and the Claude Code, Codex, Pi, or OpenCode session that started it. Open, restart, or stop a server’s whole process tree in one click, and Clean up finds the ones you forgot.
Technical approach
SwiftUI for macOS 14 and later. Everything is read locally from lsof, sysctl, nettop, and local session files. No account and no telemetry. Distributed by PyModel LLC, signed with an Apple Developer ID and notarized.
Hosted MCP · design skillActive system

Niblet

Real screen references and a design skill for coding agents that build UI.

For Teams whose coding agents ship generic interfaces and burn tokens guessing at layout.

Problem
Coding agents build UI from generic templates, so screens drift from the product and miss the states real users reach.
What PyModel built
A searchable catalogue of real iOS screens and web design packs, served to coding agents through a hosted MCP, plus an open-source design skill that works without an account.
Technical approach
A remote MCP with account keys, a REST API, and an npm package (@pymodel/niblet). The skill settles the screen’s job and states first, builds from the codebase’s own tokens, and checks the rendered result.
MCP server · open sourceActive system

Jev Judge MCP

Typed judgment tools for MCP agents, backed by TypeSafe AI’s Jev model.

For Coding agents that need a fast, independent check before they act or claim a task is done.

Problem
Agents grade their own work. A bounded question, such as whether a claim is supported or a patch finishes the task, deserves an independent answer.
What PyModel built
An MCP server that exposes Jev as verify, screen, find, classify, rerank, decide, compare, extract, review, gate, and score tools.
Technical approach
The model returns probabilities, and policy turns them into one of three actions: auto, review, or escalate. Python, published on PyPI, with one-command install for Claude Code, Codex, Cursor, and other MCP hosts.

Your system starts with a specific operating problem.

Bring the workflow, data, constraints, and outcome. PyModel will define the smallest AI and software system that can prove value.

Discuss your initiative