July 2026 Summaries

Entire’s Dispatch 0x0017 announces a fast Git hosting platform in India designed to improve local mirroring, cloning, and pulling for developers and agents, alongside updates across its CLI, web platform, regional infrastructure, and open-source projects. Entire CLI v0.9.0 improves refs-native checkpoint setup, agent guidance, diagnostics, installation instructions, and script-friendly session handling, while entire.io adds more consistent repository and commit file trees, clearer repository selection and navigation, improved code rendering, and recovery from stale frontend loads. Reliability work includes handling partial repository indexes without fatal errors, generating Dispatches from regional API cells, improving sandbox capacity management, routing account deletion and repository metadata through users’ home jurisdictions, and preserving collaborator access during onboarding. Backend changes optimize repository maintenance, improve permission synchronization and service-account credential handling, and add audited controls for blocking waitlisted accounts. Updates to go-git strengthen fetch safety by preventing silent tag movement, improve protocol validation and shallow-clone behavior, reduce unnecessary cleanup after cancellations, and incorporate security fixes in the v5 release line.

Marvin, a digital agent, reports on the latest updates and improvements to the Entire platform, which includes the launch of a fast Git hosting service in India to enhance clone and pull speeds for developers. The platform's CLI has seen updates for cleaner refs-native setup and improved guidance for agents, alongside clearer setup and diagnostics. Enhancements to Entire.io include better repository file tree navigation and reliability in dispatch generation and repository selection. The platform now offers improved settings navigation and agent run reliability, with sandbox admission regulating run deferrals and capacity outages. Code rendering has been optimized for readability, while account management and onboarding processes have been streamlined to respect regional jurisdictions and maintain shared access. Repository maintenance has been optimized for performance, with integrity checks ensuring efficient history walks and permission synchronization verifying revocations. Updates to the OSS project go-git include safer fetch operations and protocol handling, ensuring reliability and security. The dispatch encourages feedback and contributions through Discord or GitHub.

Marvin introduces Entire's in-browser code viewer, which now enables users to follow code at the token level and preview Markdown without leaving the repository, enhancing the understanding of agent-generated code. This tool allows users to trace unfamiliar symbols across different files and access related documentation to build a comprehensive mental model. By selecting a navigable token, users can view its definition and usages in a symbol panel, even in other files, provided the file has symbol information. The feature also supports Markdown previews that render README files and other documentation with functional links to related files and headings, offering a seamless exploration experience. Users can explore these features by opening a repository on Entire, navigating to the Overview, and viewing all files, with the option to try it without signing in on a public repository like the Entire CLI.

Entire’s in-browser code viewer now offers token-level code navigation, allowing users to select resolvable symbols such as function calls and follow them to their definitions and usages across repository files. The feature also preserves links to related files and documentation headings, enabling users to move between implementation details and explanatory materials without leaving the repository. Designed to help developers understand unfamiliar or agent-generated code changes, it can be tested through the Entire CLI by locating the implementation of the `entire enable` command, tracing its symbols, and comparing the code with its README documentation.

Marvin introduces a new feature in Entire that allows users to import pre-existing sessions from coding agents into their repositories, providing flexibility for those who started working before adopting Entire. During the initial setup, Entire scans for importable sessions from the past 30 days, offering users the choice to import specific agents' session histories. Users can also import sessions at any time after setup, with a recommended dry run to assess the number of turns and sessions available. Once imported, these sessions appear as read-only records with accessible prompts, responses, timestamps, model information, and token usage, though they cannot be resumed or rewound. This capability is available for all built-in agent integrations except OpenCode, and further details can be found in the Import Past Sessions section.

Entire allows users to import coding-agent sessions that occurred before the tool was enabled in a repository, either during initial setup or later on demand. During setup, it searches preferred agent integrations for sessions from the previous 30 days and lets users select which histories to import, while on-demand imports can be previewed with a dry run before completion. Imported sessions are supported for all built-in agent integrations except OpenCode and appear as read-only history records that can be searched and reviewed for prompts, responses, timestamps, model details, and token usage when available, but cannot be resumed or rewound.

Marvin provides an overview of the latest updates and enhancements to the Entire platform, emphasizing improvements in CLI usability, checkpoint management, authentication, and repository handling. Key updates include enhanced attribution for subagents, smoother repository code navigation, and more reliable checkpoint pushes, with new features such as open-sourced go-nuts for building NATS and JetStream services in Go. The platform now supports safer authentication and redaction, better session and commit filtering, and cross-region mirroring capabilities to enhance reliability. Improved user interface elements like interactive review setups and code navigation tools offer more streamlined workflows. Backup processes have become more efficient, focusing on incremental changes to minimize storage needs, while repository and account management features have been refined to improve synchronization and deletion processes. Overall, these updates aim to enhance the user experience, streamline operations, and ensure robust data handling across the platform.

Dispatch 0x0016 details broad improvements to Entire’s CLI, web platform, database infrastructure, and open-source projects, with a focus on more reliable checkpointing, agent-session capture, authentication, repository setup, and cross-region mirroring. The CLI adds structured checkpoint output, safer enablement defaults, improved first-push and SSH handling, better support for submodules and Windows, stronger secret redaction, and more accurate attribution of subagent changes and token usage across full transcripts. Entire.io improves sign-in performance, repository browsing, pinning, search, multi-filter session views, public-session routing, Markdown rendering, and code navigation that links tokens to the correct definitions and usages even where names are duplicated. EntireDB now handles default-branch renames, large repository lists, permissions, mirror creation, deletion workflows, and incremental backups more predictably while reducing unnecessary storage and processing. The release also introduces a public endpoint for outbound IP ranges, updates git-sync and ForgeMark, strengthens go-git’s ref validation, tag safety, credential redaction, worktree behavior, and submodule patches, and open-sources go-nuts, a Go toolkit for resilient NATS and JetStream services with lifecycle management, tracing, retries, reliable publishing, and dead-letter support.

Ref-based checkpoint storage is a new feature in Entire that optimizes the storage of checkpoints by assigning each checkpoint its own Git ref, rather than using a shared branch. This change enhances performance by allowing checkpoints to be saved and accessed independently, which is particularly beneficial as the agent history expands. It also enables multiple agents to save simultaneously without conflicts. The transition to this storage method is seamless and does not alter the existing workflow, allowing users to continue creating, searching, explaining, and resuming checkpoints with the same commands. Available in Entire CLI version 0.8.42 and later, ref-based storage can be enabled in new or existing repositories through specific commands, and it features longer, time-sortable checkpoint IDs compared to the branch-based system.

Entire has introduced an opt-in ref-based checkpoint storage system in CLI version 0.8.42 and later that assigns each agent checkpoint its own Git ref rather than storing all checkpoints on the shared entire/checkpoints/v1 branch. This approach is intended to preserve fast checkpoint saving and retrieval as agent history grows while allowing multiple agents to save work concurrently without contention on a shared branch. Existing Entire workflows and commands for creating, searching, explaining, and resuming checkpoints remain unchanged, though ref-based storage uses longer time-sortable checkpoint IDs. Users can enable the option with `entire enable --checkpoint-backend refs` for new repositories or `entire configure --checkpoint-backend refs` for repositories already configured with Entire.

Entire has introduced `entire agent-help`, a CLI command that gives coding agents live, version-specific guidance for using the Entire CLI. Designed to replace frequently outdated command documentation embedded in agent Skills, it provides a map of available capabilities, links common tasks such as searching checkpoints, creating dispatches, reviewing branches, and managing integrations to relevant commands, and allows agents to inspect exact command options when needed. The command detects the current repository and limits its guidance to supported features, reducing unnecessary queries and unusable commands while preserving agent context for coding work. Available in Entire CLI version 0.8.42 and later, it can be run in Entire-enabled repositories and can return structured JSON output through the `--json` option.

Entire has introduced a new CLI command called "entire agent-help" designed to provide coding agents with direct guidance on the Entire CLI, ensuring the guidance always matches the installed version. This command moves the guidance into the CLI itself, offering agents context to select the appropriate command and look up details as necessary, which reduces the need for frequent updates to CLI guidance within Skills and allows Skills to focus more on guiding workflows. When executed, "entire agent-help" generates a live map of the CLI, helping agents navigate tasks like searching past checkpoints or managing integrations without spending extra tokens or encountering irrelevant commands. The command also tailors its output to the current repository, displaying only applicable features, and supports structured output in JSON for agents and tools. Users can access this feature by updating to Entire CLI 0.8.42 or later.

Entire’s Dispatch 0x0015 announces the preview launch of its distributed Git network, which lets developers mirror GitHub repositories across Australian, US, and EU regions to provide faster local access for agents, alongside the open-source ForgeMark benchmark used to test concurrent Git workloads. The company reports throughput of up to 25 times competing platforms and roughly 570,000 clones per hour from one repository, while introducing extensive CLI, web, storage, and infrastructure improvements. Key changes include regional authentication reuse and automatic refresh, expanded mirror-management commands, imported agent-session support, agent-oriented CLI help, and an opt-in ref-based checkpoint model intended to improve push speed and reduce read costs. The web platform adds mirror-status visibility, regional controls, improved repository browsing, run metadata, recovery options, and more resilient onboarding, while EntireDB receives performance, replication, security, storage-integrity, and job-reliability upgrades. Open-source projects also advanced through git-sync support for Entire URLs and finer branch exclusions, ForgeMark portability and CI improvements, and go-git’s adoption of Git Protocol v2 by default, hook support, and faster tree reading. Finally, Entire updated its Terms of Service, Privacy Policy, Cookie Policy, and introduced a Data Processing Agreement.

Marvin's latest update highlights the launch of Entire's distributed Git network, designed to enhance Git hosting with faster and more reliable access through regional mirroring in Australia, the US, and the EU. The platform's performance was tested using ForgeMark, an open-source Git push benchmark, showing significantly higher throughput compared to competitors. Key updates include improved CLI capabilities, such as enhanced mirror operations and regional authentication, a new ref-based checkpoint storage model for efficient data management, and various UX enhancements to streamline user interaction. Additional developments encompass improvements in repository mirroring, storage resilience, onboarding processes, and integration with OSS projects like git-sync and go-git. Entire's Terms of Service and related policies have been updated to reflect its evolving business processes, and the company encourages feedback through Discord or GitHub.

Entire has introduced a distributed, Git-compatible repository network designed to reduce the rate-limit and concurrency challenges created by coding agents that generate high volumes of Git traffic. Users can mirror public or private GitHub repositories into Entire’s regional cells in the United States, Europe, Australia, and, following an update, India, allowing agents to clone and fetch locally while GitHub remains the origin; GitHub-backed pushes remain limited by GitHub’s performance, while Entire-native branches support higher-concurrency workloads. Its architecture separates a global control plane for identity and placement from regional data planes that store content-addressed Git data, with replication across availability zones and options for region-specific storage or multi-region redundancy. Entire also released ForgeMark, an MIT-licensed benchmarking tool for measuring concurrent Git push performance under agent-like workloads, and offers special unmirrored branches for direct regional write performance. Access is initially available to existing users, with new users placed on a waitlist, while the company plans to add native repository hosting, open-source and self-hosted backend support, verifiable branch protections, CI/CD capabilities, and enterprise management features.

Git was originally designed as a decentralized version control system, but modern hosting practices have led to a more centralized approach. To address this, Entire offers a solution that mirrors public or private GitHub repositories across a distributed network, allowing for faster cloning and reduced load on the origin repositories. Entire's network uses regional nodes to handle concurrent requests efficiently, providing a scalable, low-latency architecture that is optimized for agent-based workloads. The system ensures data availability through redundancy and regional control over repository content. Entire plans to further decentralize by open sourcing its network, enabling self-hosting, and expanding globally, with the ultimate goal of fostering a collaborative, decentralized development environment. Users can start by installing the Entire GitHub App and CLI, and mirroring their repositories to benefit from this distributed architecture.

After more than two decades of widespread adoption, Git remains the dominant version control system, but the rise of AI-generated code may require Git hosting to evolve beyond recording code changes and commit messages. The post argues that agent session logs, including prompts, tool calls, decisions, and checkpoints, should be stored alongside code as a semantic memory layer that preserves development intent, helps agents avoid repeated mistakes, reduces review time, and improves human verification. It also contends that Git’s decentralized design should be more fully realized, since centralized hosting platforms could face rate limits, outages, and scalability constraints as large numbers of agents work simultaneously. A distributed network of repository hosts could improve resilience, local access, data residency, and organizational control while enabling global collaboration. Under this vision, Git would remain the source of truth for software, coordinating parallel work among humans and agent fleets while preserving code provenance.

Git, the most popular version control system for over two decades, faces a need to evolve as AI agents increasingly produce code, necessitating a shift in how code and its context are stored and managed. The current centralized hosting platforms may not sustain the rapid pace and parallelization of agent-driven development, prompting a return to Git's decentralized roots where repositories are widely distributed. This change would enhance availability, resilience, and digital sovereignty, allowing developers and organizations to maintain data residency while participating globally. The proposed addition of a semantic memory layer, capturing session logs alongside code, aims to improve accuracy, productivity, and understanding by preserving the context of development decisions, thus facilitating better collaboration between humans and agents in building software efficiently.