brethof-brain
A system that provides long-term memory for AI agents. Through three components for writing, retrieval and injection, it automatically generates session briefings, replacing the workflow of repeatedly pasting prompts and allowing agents to retain rules, decisions and project context across sessions.
Timeline (last 90 days)
Since December 2025, the developer of brethof-brain's team's self-developed Agent began running on a set of memory system, namely the later-productized brethof-brain; since then, preambles are no longer pasted.
brethof-brain's rule set is limited to within 1000 characters each, and the memory system will point out duplicates and remind whether the rule set is too large, but will never edit or delete on the user's behalf.
The steps to install brethof-brain are: tell the Agent 'install brethof-brain', and it will pull the client, set up the hooks, and self-check; the manual steps are only registration, choosing local or hosted, and approving the connection on brethof.ai when the installer requests it.
In this system, projects are isolated from one another, each with its own records, rule graphs, and plans, and rules have a capacity cap that triggers a warning at 80%.
The memory system is implemented in three layers: curated records (current truth), a decision graph (recording the decision-making process), and complete history (all content, searchable).
The article is essentially a product promotional piece; the end of the article provides the installation instruction 'install brethof-brain', the official website link, and an affiliate promotion program, whose operational claims lack independent verification.
Memory in this system is stored on the user's local machine (local container) or in encrypted cloud storage, and the hub that handles briefings retains no fragments.
The curated records system uses a window of every five conversation turns to read content, retaining what was finalized in the conversation and reconciling new and old facts.
The decision graph uses an append-only model, with one line per decision; overturning a decision adds a new line, and history is never rewritten.
All Facts (9)
brethof-brain's rule set is limited to within 1000 characters each, and the memory system will point out duplicates and remind whether the rule set is too large, but will never edit or delete on the user's behalf.
50%UnverifiedThe curated records system uses a window of every five conversation turns to read content, retaining what was finalized in the conversation and reconciling new and old facts.
50%UnverifiedThe decision graph uses an append-only model, with one line per decision; overturning a decision adds a new line, and history is never rewritten.
50%UnverifiedIn this system, projects are isolated from one another, each with its own records, rule graphs, and plans, and rules have a capacity cap that triggers a warning at 80%.
50%UnverifiedMemory in this system is stored on the user's local machine (local container) or in encrypted cloud storage, and the hub that handles briefings retains no fragments.
50%UnverifiedThe article is essentially a product promotional piece; the end of the article provides the installation instruction 'install brethof-brain', the official website link, and an affiliate promotion program, whose operational claims lack independent verification.
50%UnverifiedThe memory system is implemented in three layers: curated records (current truth), a decision graph (recording the decision-making process), and complete history (all content, searchable).
50%UnverifiedSince December 2025, the developer of brethof-brain's team's self-developed Agent began running on a set of memory system, namely the later-productized brethof-brain; since then, preambles are no longer pasted.
50%UnverifiedThe steps to install brethof-brain are: tell the Agent 'install brethof-brain', and it will pull the client, set up the hooks, and self-check; the manual steps are only registration, choosing local or hosted, and approving the connection on brethof.ai when the installer requests it.
50%