Doka launches local AI-agent for Russia without VPN, subscription or cloud

Russian developer presented Doka — a local AI agent designed to address the main barrier for Russian users: dependence on VPN, foreign subscriptions, and cloud infrastructure. The application works as a desktop assistant for web research and everyday tasks, with the base model running directly on the device.

From Personal Tool to Product

The idea for Doka grew from simple frustration. Over the past year and a half, tools like Cursor, Claude Code, and Copilot have noticeably accelerated developer workflows, but in Russian reality they often run into unstable VPN, complex payment systems, and the risk of sudden restrictions. The project author decided to understand how agentic systems work under the hood: he studied agentic loops, tool calling, and schedulers, then built his first working prototype.

Initially, it was purely an internal tool without a UI or onboarding. The agent could search for information online, open pages, read their content, and show the work step-by-step: from search query to final answer. Only the author could use it because without understanding the internal logic, the tool was almost impenetrable for a newcomer.

The turning point came after a live demonstration to a friend: it became clear that the value here wasn't in the engineering experiment, but in a convenient wrapper for an ordinary user.

"What even is this? Can I download it?"

What Doka Can Do

The current version of Doka is intentionally narrow in scope. The application focuses on internet search and web page interaction: a user formulates a task, and the agent selects queries, reads documentation or articles, and returns a structured result. Essentially, it's an agentic interface for everyday web research. The author uses it for preliminary research before tasks, morning monitoring of news in ML and backend, as well as for parsing long and poorly structured documentation.

• Search and read web pages without manually switching between tabs
• Brief breakdowns of unfamiliar technologies, libraries, and architectural approaches
• Daily digests on given topics like ML and backend
• Help with lengthy documentation that's easier to delegate to an agent
• Completion of routine research tasks that would otherwise take an hour to do manually

The key bet is on a local model. By default, Doka installs Qwen3, and according to the author, this is enough for most real scenarios involving search, reading, and explaining page content. This isn't an attempt to compete with top cloud models in absolute quality, but a pragmatic choice for accessibility. Web tasks still require access to websites, but the model itself runs locally, doesn't send user queries to the cloud, and doesn't require monthly payments.

Where the Project Is Heading

The author describes the current version as a minimal starting configuration needed not to maximize features, but to test demand and use cases. That's why Doka's development is planned in stages, from the most practical capabilities to more complex ones. The priority is not abstract "universality," but a set of features that save time right now. This approach should help avoid spreading thin on decorative features and understand faster what users will actually come back for.

The roadmap already outlines work with local files, screenshot and current screen analysis, long-term memory between sessions, and background tasks that can be launched and left running parallel to the main work. The application is already available for free, supports Windows 10/11 and macOS 12+, and the model downloads automatically on first launch based on suitable hardware. An important detail: users don't need to go into the terminal or manually configure the environment.

What This Means

Doka shows how rapidly the AI tools market is changing: the focus is shifting from universal cloud services to local agents adapted to specific user constraints and preferences. For the Russian market, this is a particularly important signal — demand exists not only for the "smartest" model, but for a product that works stably without VPN, a foreign card, and privacy compromises. If the trend takes hold, local assistants will become a separate product category rather than a niche experiment.