Sketch offers a container-native agent development environment that isolates LLMs in dedicated Docker containers for autonomous tool management and workflow execution. It captures all agent actions as Git commits, providing traceability and a web UI for engineers to review and guide LLM-driven code changes.
Funding
Funding not disclosed
Founders
Product
Problem
Traditional software development workflows struggle to integrate Large Language Models (LLMs) effectively, as LLMs require dedicated environments for autonomous operation and tool management. This lack of a structured, isolated environment hinders LLM agency and complicates the tracking of their development processes.
Solution
Sketch provides a container-native agent development environment that isolates LLMs in dedicated Docker containers, granting them autonomy to manage tools and development workflows. The platform captures all agent actions and changes as standard Git commits, ensuring traceability and version control. A web-based UI offers a code review interface for engineers to interact with the LLM, review proposed changes, and guide the development process. This approach separates LLM operations from human developer environments, enabling LLMs to install their own dependencies and operate without compromising production systems.
Target Audience
The primary users are software engineers and AI developers engaged in building and managing autonomous AI agents for complex, long-lived software projects.
Features
- Isolated Docker containers for each LLM agent session, ensuring environment separation.
- Autonomous tool management and dependency installation (e.g., `apt-get install`) within agent containers.
- Git commit generation for all agent-initiated code changes, providing a clear audit trail.
- Web-based UI for reviewing code diffs, agent thought processes, and facilitating code review.
- Support for local execution with user-provided LLM models for enhanced data privacy.
- Cloud-hosted option with GitHub integration for streamlined development workflows.
- SDKs and APIs for programmatic interaction and integration with existing CI/CD pipelines.
- Browser automation tools for agents to interact with web interfaces.