Kontinuo is a specialized continuity layer designed for AI coding agents, providing a robust mechanism to maintain context and intent across different agent sessions. By creating verifiable checkpoints, it allows one AI agent to leave a clear record of its progress, which a subsequent agent can then read and utilize to resume tasks seamlessly. Developed to operate as a local-first utility, the platform ensures that no session data, code, or context leaves the user's machine, maintaining privacy and security by design. It bridges the gap between disconnected agent sessions, effectively eliminating the need for developers to manually re-explain project status or previous progress when switching between different coding agents or environments.

Functionality is centered on the ability to capture, verify, and serve handoff notes that are intelligible to both humans and AI models. Kontinuo functions by observing agent sessions, Agent Client Protocol (ACP) streams, and git workspace states to generate a normalized checkpoint for the current repository. Each checkpoint includes critical information such as the current project goal, the precise stopping point, pending tasks, and a workspace fingerprint. Before a new agent begins its work, it can use the Model Context Protocol (MCP) to ingest these checkpoints, ensuring it understands exactly where the previous agent left off, which files were modified, and what the next logical step in the development process should be.

Some of the key features are:

• Verifiable Checkpoints: Records evidence including git HEAD, changed files, dirty state, and workspace fingerprints to detect stale state before an agent acts.
• MCP Integration: Uses the Model Context Protocol to serve handoffs, allowing MCP-capable agents to read and write state without needing to parse external, proprietary logs.
• Local-First Architecture: Operates entirely on the local machine with no cloud accounts required, ensuring all session data and code remain private.
• Protocol-Level Capture: Utilizes ACP support to observe traffic directly between editors and agents for clearer intent capture.
• Standardized Schema: Employs an open JSON schema for checkpoints, ensuring compatibility across different agent hosts and tools.
• Secret Redaction: Automatically scrubs sensitive information to ensure that token formats or private data are not persisted in handoff notes.

Operation is facilitated through a simple command-line interface (CLI) and an MCP server that integrates with existing agent host configurations. Users install the binary locally, register their specific agent host (such as Claude Code, Cursor, or OMP) with the workspace, and the tool then handles the capture of agent activity in the background. As the agent completes tasks, Kontinuo records the state, which is then made available for retrieval by the next agent upon initialization. This creates a loop of verifiable progress where the agent acts based on the latest, verified checkpoint, reducing redundant efforts and context loss.

Some common use cases include:

• Long-term Refactoring: Allowing an AI agent to handle large, multi-session refactoring tasks where the project state changes significantly between sessions.
• Tool Switching: Enabling developers to switch between different AI coding agents (e.g., from Claude to Codex) without losing context regarding what was just completed or what is planned next.
• Mid-Session Handoffs: Managing transitions when an agent hits a context window limit, allowing a new session to pick up exactly where the last one was forced to stop.
• Parallel Development Streams: Maintaining clarity on workspace changes and pending tasks when multiple agents or sessions interact with the same code repository over time.