March 2026 Summaries

Marvin introduces the Entire CLI, designed to enhance the understanding of software changes by binding agent context, prompts, and chat logs to git. This week, the CLI expanded its capabilities with the introduction of three new agents—OpenAI Codex, Amazon's Kiro, and Pi—through a new plugin system, alongside the addition of Windows support and performance improvements for large Checkpoint histories. OpenAI Codex is now available in preview, offering lifecycle hooks and transcript parsing, while Kiro and Pi are accessible via external plugins. The update also includes bug fixes and enhancements to the Entire web platform, such as improved tool call rendering and streamlined Checkpoint metadata performance. Additionally, users can now configure remote repositories for Checkpoint histories and address issues with fresh clones and bare repositories. The entire.io website has been optimized for AI interaction, enhancing accessibility and user experience.

Entire CLI version 0.5.2 expands its AI coding-agent integrations by adding OpenAI Codex in preview and plugin-based support for Amazon Kiro and Pi, while continuing to record agent context, prompts, and chat logs alongside Git activity through Checkpoints. The release also introduces Windows compatibility, performance improvements for large Checkpoint histories through more efficient metadata retrieval, configurable dedicated remote repositories for Checkpoint data, and fixes for fresh clones, bare repositories, and empty repositories. Additional updates address issues across Cursor, Gemini, Copilot CLI, OpenCode, and hook-manager detection. On the web platform, tool calls have been redesigned for clearer inspection, navigation and checkpoint highlighting have been improved, and the website now supports Markdown rendering, RSS, accessible HTML, and reader-friendly content delivery.

Entire introduces Checkpoints, addressable records that preserve the context behind agent-assisted code changes, including code state, prompts, agent transcripts, token usage, file changes, and line-level attribution between agent output and human edits. Integrated into existing Git workflows through the Entire CLI, Checkpoints are linked to commits via persistent trailer IDs that survive rebases, amends, squashes, and cherry-picks, while their metadata remains separate from working branches. The system stores data as Git objects, either alongside code or in a separate repository for privacy and management needs, and uses temporary records during active work before creating permanent records upon commit. By retaining the intent and process behind changes, Entire aims to improve team visibility, searchable development history, auditing, incident investigation, and code review, enabling reviewers to assess both a change’s implementation and the agent-driven reasoning that produced it.

In 2026, developers face the challenge of managing the vast amount of code produced by AI agents, leading to the introduction of "Checkpoints" to capture not only code changes but also the context behind them. Checkpoints bundle the code state, agent transcripts, prompts, token usage, and line-level attribution, allowing developers to trace changes back to their origins. This system integrates seamlessly with existing git workflows, storing metadata in git objects and utilizing a two-tier storage model to manage temporary and permanent data. The Entire CLI preserves session history without altering the traditional git experience, supporting multiple agents and allowing for concurrent sessions. As development shifts towards AI-driven workflows, Checkpoints aim to provide deeper insights into the "why" behind code changes, enhancing team visibility, searchability, and auditability. This evolution in code management is set to transform code reviews into intent reviews, focusing on problem-solving and decision-making processes rather than just syntax, ultimately paving the way for a new developer platform.

Marvin's latest dispatch reveals enhancements to the Entire CLI, focusing on performance improvements and bug fixes, particularly for users working with large repositories. Key updates include significant memory usage reductions through sparse metadata fetching and new content detection methods, as well as upgrading to go-git/v6 for enhanced efficiency. Users can now store checkpoints in a separate repository, maintaining a clean working repo history while ensuring full access to checkpoints across clones and teams. The update also introduces automatic discovery of external agent plugins, fixes for Gemini CLI hook validation, and enhancements to the entire rewind function to properly resolve file paths from subdirectories. Additionally, a new diagnostic command, entire trace, is introduced for identifying performance issues, and the entire update function now accurately detects installations via mise. The dispatch concludes with a call for feedback and an open invitation for questions on Discord.

Entire CLI’s Dispatch 0x0006 announces ongoing weekly updates focused on building an independent semantic layer that captures development context across Git commits and AI agents. Key improvements include reduced memory use in large repositories through sparse metadata fetching, byte-based content detection, and an upgrade to go-git/v6 with LazyIndex support. Users can now store checkpoint branches in separate remote repositories, while checkpoint history displays all branch commits alongside captured-context checkpoints. The release also adds the entire trace diagnostic command for identifying performance bottlenecks, fixes rewind behavior from repository subdirectories, improves mise installation detection during updates, and enhances compatibility and automatic discovery for external agent plugins, Gemini CLI, Cursor, and Factory.

A remote-first company held an in-person Hack Week for regional teams in Melbourne, Seattle, and Lisbon, bringing together 25 employees to collaborate on projects and ideas. The update also announced support for the GitHub Copilot CLI, including checkpoints, session resumption and rewind capabilities, transcripts, and model tracking, with Entire positioned as a tool for recording and sharing AI-assisted coding workflows and connecting developer intent to resulting code.

Entiron conducts bi-annual in-person Hack Weeks to foster collaboration among its remote-first teams, with the latest gatherings held in Melbourne, Seattle, and Lisbon, involving 25 participants who engaged in building and idea-sharing activities. A notable development from this week is the integration of GitHub Copilot CLI into their operations, which now features checkpoints, resume, rewind, transcripts, and model tracking capabilities, allowing users to seamlessly share AI coding sessions and link intentions to code outcomes. The company encourages continued contributions and feedback from its community, hinting at further enhancements and updates in the near future.

Entire CLI’s latest update expands its portable agentic context platform, which automatically syncs chat sessions, logs, and prompts to Git repositories, through a new Factory integration, an organization-wide Repositories section on entire.io, and a plugin system that lets third-party agents operate as standalone programs using a common protocol. Interface improvements add recently used and pinned repositories, more compact checkpoint views, and breadcrumb navigation between commits. Performance work replaces slower Go-based repository scans with native Git operations, adds timing instrumentation, and caches staged-file information to reduce hook latency in large repositories. The release also improves Cursor transcript analysis and session metrics, ensures OpenCode sessions close correctly, adds Factory end-to-end testing, strengthens squash-merge resume and rewind behavior, records LLM models in checkpoint metadata, resolves stale-session and file-tracking issues, and introduces an absolute Git-hook path option for more reliable setup across shells and CI environments.

Dispatch 0x0004 highlights recent updates to the Entire CLI, focusing on improving integration, performance, and user experience. Key enhancements include a new Factory AI integration, an external agent plugin system for seamless third-party agent integration, and user interface updates for better navigation and access to repositories. Performance improvements were made to post-commit behaviors, particularly for larger repositories, by utilizing native git operations and caching strategies. Updates to the Cursor, Factory, and OpenCode agents have enhanced session metrics and transcript analysis, while new E2E test runners and fixes have improved session reliability. The system now also tracks which large language model (LLM) was used in checkpoints, improving traceability. Developers benefit from improvements in setup, such as support for absolute git hook paths and fixes to commit signing issues in the test suite. Overall, these updates aim to provide a more efficient and integrated experience for users collaborating with AI companions.