The Pulse
Entire uses saved coding sessions to address agent memory gaps
Entire says its Continuous Memory system lets coding agents retrieve decisions from earlier sessions and check their sources. A company test showed a gain on one task, but its benchmark documentation limits the result to one repository and a specific test setup.

Entire is using saved coding-agent sessions and repository history to give agents access to decisions from earlier work, rather than relying only on summaries of past conversations. The company described the approach, called Continuous Memory, in a November 18 post.
Its system combines two open-source, local-first plugins for the Entire command-line interface: entire-graph maps relationships in code, while entire-brain connects that map to saved sessions and other project material. The aim is to let an agent retrieve earlier decisions and inspect their sources before starting a task.
What gets lost in a summary
When a long session fills an agent’s context window, a tool may compact the conversation by replacing older messages with a summary. That is separate from memory that carries information between sessions.
Entire argues that summaries can omit details or introduce things that did not happen, and that repeated compaction can carry mistakes forward while losing more context. A larger context window can make summarization less frequent, the company says, but does not change the fact that the agent eventually works from a summary.
Continuous Memory instead draws on sessions saved in the repository. It does not replace instruction files such as CLAUDE.md and AGENTS.md; setup adds references to those files so agents know when to use the memory tools.
How the plugins work
Entire records sessions as-is in Git and links them to commits through checkpoints. Supported agents include Claude Code, Codex, Cursor, Copilot CLI, Gemini CLI, OpenCode, Factory Droid and Pi; other agents can be added through an external agent plugin.
The brain plugin indexes saved sessions and attaches them to the code graph. It also reads Git history and documents, including PDF, Word, Excel and PowerPoint files in the repository’s docs folder.
Extracted facts link back to a session, commit or file, and the entire brain verify command checks whether those links still match the code. A brief command supplies an agent with a packet of relevant decisions, previous attempts, code and tests before a task begins.
The project memory can also be used when switching agents, according to Entire’s documentation. The plugin can work from code, documents and Git history even if no saved sessions are available.
Test results and operating limits
In a company test, an agent was given a task whose answer was in an old session rather than the code. Entire says it passed 0 of 5 runs without entire-brain and 5 of 5 with it, retrieving the earlier decision from history.
The benchmark documentation says that result covers one task in one repository, with five runs per condition, and is limited to the specified runner and history channel. A second task passed in all 5 runs under both conditions, which the documentation says is not evidence of an improvement. The company says it has not run a defensible head-to-head test against another memory product.
Memories remain on the user’s machine by default; syncing begins when a repository is connected to its hosted Entire repository. Extracting memories from past sessions uses an available coding agent, can send session content to its model provider and can add token charges. Setup asks before extracting those memories, with no as the default answer.