Slowave
An open-source, locally run memory layer tool for programming agents. It dynamically maintains memory salience through reinforcement and decay mechanisms, enabling agents to participate autonomously in memory management. Data is stored via SQLite and services are exposed through MCP endpoints.
Timeline (last 90 days)
Slowave is a fully locally-running open-source tool serving as a programming agent memory layer, using a lightweight multilingual embedding model with data stored in SQLite, requiring no external memory services or LLM APIs
Slowave allows programming agents to participate in maintaining their own memory, adjusting memory salience through agent feedback signals (useful, irrelevant, or outdated)
Slowave introduces a mechanism similar to biological memory: useful memories are reinforced, useless memories decay over time, and the rest are treated as noise.
Slowave's memory cycle is: remember → recall → use → feedback → reinforce/weaken → decay.
Slowave works through 5 MCP endpoints.
Clients supported by Slowave include Claude Code, Codex, Cursor, Cline, OpenCode, Windsurf, and Claude Desktop.
Slowave has been open-sourced on GitHub, with the repository at slowave-ai/slowave.
Slowave's core assumption is that retrieval is only part of the memory problem, and what truly matters is whether memories that are helpful to the current task and goal can be retrieved.
Because the agent itself evaluates whether a memory is useful, Slowave no longer needs a separate LLM judge to maintain memories, thereby avoiding token waste and the split-brain problem.
Slowave's decay mechanism may carry the risk of accidentally deleting memories that are truly important but temporarily unused, and the accuracy of the agent's self-evaluation of memory usefulness also needs to be verified through real-world testing.
All Facts (10)
Slowave's decay mechanism may carry the risk of accidentally deleting memories that are truly important but temporarily unused, and the accuracy of the agent's self-evaluation of memory usefulness also needs to be verified through real-world testing.
50%UnverifiedSlowave allows programming agents to participate in maintaining their own memory, adjusting memory salience through agent feedback signals (useful, irrelevant, or outdated)
50%UnverifiedSlowave introduces a mechanism similar to biological memory: useful memories are reinforced, useless memories decay over time, and the rest are treated as noise.
50%UnverifiedSlowave's memory cycle is: remember → recall → use → feedback → reinforce/weaken → decay.
50%UnverifiedSlowave works through 5 MCP endpoints.
50%UnverifiedClients supported by Slowave include Claude Code, Codex, Cursor, Cline, OpenCode, Windsurf, and Claude Desktop.
50%UnverifiedSlowave is a fully locally-running open-source tool serving as a programming agent memory layer, using a lightweight multilingual embedding model with data stored in SQLite, requiring no external memory services or LLM APIs
50%UnverifiedSlowave's core assumption is that retrieval is only part of the memory problem, and what truly matters is whether memories that are helpful to the current task and goal can be retrieved.
50%UnverifiedBecause the agent itself evaluates whether a memory is useful, Slowave no longer needs a separate LLM judge to maintain memories, thereby avoiding token waste and the split-brain problem.
50%UnverifiedSlowave has been open-sourced on GitHub, with the repository at slowave-ai/slowave.
50%