Anton Razzhigaev & Andrew Kaznacheev
Anton Razzhigaev and Andrew Kaznacheev are individual creators/researchers rather than a separately incorporated company. Their principal project, Ouroboros, is an open-source general-purpose AI coding agent with persistent identity, memory and history that can coordinate specialist agents and evolve its own code, prompts, tools, architecture and dependencies; Claudexor is the local-first control plane used for connected coding-agent execution.
At a Glance
- Individual developers and researchers
- Open-source software builders
- Teams needing auditable autonomous coding and verification workflows
- Users of Claude Code, Codex, Cursor, OpenCode and related coding-agent harnesses
- +1 more
AI Tools by Anton Razzhigaev & Andrew Kaznacheev
(2)
Claudexor
Local Control Plane for AI Agents
Ouroboros App
Self Modifying AI Agent Desktop
Discussions
No discussions yet
Be the first to start a discussion about Anton Razzhigaev & Andrew Kaznacheev
Latest News
Ouroboros v7.8.0 released with Windows tray/console repairs, native Linux window, managed Claudexor 4.6.2 and macOS/Windows/Linux/Android artifacts
Ouroboros v7.6.0 introduced authored memory/journal behavior, unified interface language, background startup options and task pause/resume/continue
Ouroboros v7.0.0 introduced a modular runtime, delegated inputs and visible owner dialogue, with Claudexor pinned to 3.10.1
Technical report published: Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution
Products & Services
Open-source, general-purpose self-developing AI agent. It works on external projects, maintains durable identity and memory across tasks and restarts, coordinates specialist-agent swarms, and can change its own code, prompts, tools, architecture and dependencies through reviewed Git history. Distributed as a desktop app and headless CLI.
MIT-licensed local-first control plane for coding agents. It connects Claude Code, Codex, Cursor, OpenCode and other supported harnesses, tracks subscription quota, optionally rotates profiles after confirmed limits, carries bounded task context between harnesses, supports best-of-N and cross-family review, and records typed execution evidence.
Market Position
Ouroboros positions itself as a self-developing, persistent coding agent rather than a conventional stateless coding assistant: it combines external-project execution, a live specialist swarm, durable identity/memory and reviewed changes to its own runtime. Its official site compares reported benchmark results with Codex, Claude Code, Cursor and Hermes. Claudexor differentiates from driving a single bare harness by adding multi-harness routing, quota-aware profile rotation, bounded context transfer, best-of-N races, cross-family review and typed evidence receipts.
Leadership
Founders
Anton Razzhigaev
AI/deep-learning researcher with a PhD from Skolkovo Institute of Science and Technology (Skoltech); his public profile describes him as building autonomous agents and multi-agent coding tools. Earlier work includes Kandinsky 2, OmniFusion, LLM-Microscope and the ACL 2024 paper 'Your Transformer is Secretly Linear'.
Andrew Kaznacheev
Co-creator of Ouroboros, credited alongside Anton Razzhigaev in the project's official GitHub repository and listed as a co-author of the Ouroboros technical report. His public LinkedIn profile identifies him as based in Dubai; the repository history also credits his contributions to file editing, skills, MCP, widgets and Windows signing/release work.
Executive Team
Anton Razzhigaev
Lead Creator / AI Researcher
AI Researcher, PhD candidate at Skoltech. Previously Research Engineer at Huawei (NLP Lab). Specialist in transformers and adversarial attacks.
Andrew Kaznacheev
Co-Creator
Co-creator based in Dubai, UAE. Professional in product and engineering.
Founding Story
Ouroboros first booted on February 16, 2026. The preserved first-generation account says Razzhigaev built it as an experiment in a system that creates and evolves itself; the early version ran in Google Colab, communicated through Telegram, stored memory on Google Drive and kept its implementation in GitHub. The current project carries that identity into a native desktop and headless runtime, with reviewed changes becoming part of the agent's continuing implementation.
Business Model
Revenue Model
The projects are presented as open-source, local-first software: Ouroboros is distributed under MIT terms and Claudexor is described as MIT-licensed, with runtime/model access configured by the user through connected subscriptions, remote provider APIs or local models. No paid subscription, usage price or commercial revenue model was identified in the reviewed sources.
Target Markets
- Individual developers and researchers
- Open-source software builders
- Teams needing auditable autonomous coding and verification workflows
- Users of Claude Code, Codex, Cursor, OpenCode and related coding-agent harnesses
- Operators who prefer local-first software and control of credentials, files and task history
- Autonomous software development and repository maintenance
- Research, debugging, implementation and verification in external workspaces
- Multi-agent coding workflows with planning, independent builds and adversarial review
- Long-running agent work that needs durable memory, identity, task history and restart continuity
- Coding-agent subscription routing and quota-aware execution across multiple harnesses
- Local-first, auditable best-of-N candidate generation and cross-model review