The system is evolving.
Follow the latest deployments.
A major milestone introducing direct graphical screen understanding and native execution isolation on Linux systems.
The agent now captures and processes virtual display framebuffers using YOLO-style models to locate and interact with graphical UI components directly.
Natively packaged sandbox build with bundle-level dependencies, bringing zero-setup editor execution to all major Linux distributions.
Integrate custom API keys directly with Anthropic, OpenAI, and Google Gemini, eliminating middleman proxy latency.
Explicit control over persistent file structures in multi-agent workflows to minimize LLM token overhead.
Performance refinements replacing Node.js execution layers with high-efficiency multi-threaded Rust scheduler logic.
Replaced task scheduling loops with native Rust threads, yielding up to 6x faster state assessment times.
Spin up dedicated headless virtual frames for GUI testing, letting the AI interact with Chrome/Firefox without stealing your mouse.
Generate visual state histories and traceback logs upon execution crash, ready to copy straight into support tickets.
Calibrated virtual mouse movements to solve alignment offsets on hidpi displays.
Opening up ZeroSphere to standard Model Context Protocol servers to dynamically load custom developer utilities.
Connect third-party tool servers (Postgres, GitHub, Slack) to expand the AI agent's capability workspace dynamically.
New planning stage displaying structured workflow steps and user consent gates before editing system files.
Optimized vector embeddings for large codebases with fast incremental rebuild triggers on file modifications.
Interactive sub-processes allowing the agent to manage ongoing background terminal tasks smoothly.