GAM (General Agentic Memory)

Use it when

• An agent's long-horizon trajectory of reasoning steps and tool invocation logs overflows the context window and must be compressed into structured, queryable memory
• You need hierarchical memory with summaries over long documents or videos so an LLM can answer questions without re-reading the full source each session

What it solves

Not the fit when

• plain vector-database similarity search; GAM organizes memory as an agent file system with LLM-generated hierarchy
• short single-turn prompts that already fit in context
• fully offline use; memory building and QA call an LLM backend
• embedding-recall-poor
• research-source-fragmentation
• knowledge-signal-noise

Install

git clone https://github.com/vectorspacelab/general-agentic-memory && pip install -e ".[all]"

Invoke

Python: wf = Workflow("text", gam_dir="./my_gam", model="gpt-4o-mini", api_key=...); wf.add(input_file="paper.pdf"); wf.request("question"), or CLI: gam-add --type text --gam-dir ./my_gam --input paper.pdf then gam-request --question "..."

Alternatives

No reviewed alternatives recorded yet.