AFS — Memory Lake for Agentic AI

The memory layer AI agents actually want to use.

AFS (Agent File System) is a CLI-first, filesystem-first memory system for agentic AI. It gives your agents a persistent, shareable, graph‑enabled memory substrate so discoveries survive across sessions, teams, and workflows.

The Problem: Agent Amnesia

Modern agents forget. Sessions end, token budgets truncate context, and parallel agents can’t share what they learn. The result is wasted time and repeated work.

WITHOUT AFS
Day 1: Agent learns "API has rate limit 1000/hr"
Day 2: Same agent asks "What's the rate limit?" → "I don't know"
Agent-1 finds bug → Session ends → Agent-2 starts fresh

Why Filesystem‑First?

Local‑first: keep memory with your codebase, no cloud dependency.
Auditable: memory lives as files you can review and back up.
Portable: move or version memory alongside your repo.
Composable: works with any orchestration framework.

How AFS Works

Three‑Tier Memory (Automatic)

AFS manages memory like human cognition—no manual promotion needed.

WORKING → EPISODIC → SEMANTIC
recent observations → full history → consolidated knowledge

Knowledge Graph (Auto‑Discovered)

AFS automatically links related memories using edges like similar_to, co_occurred, and depends_on so agents can recall context by relationship, not just keyword.

Share discoveries with a swarm so parallel agents build on each other instead of duplicating work.

Proof Points

Metric	Value
Search latency	< 100ms (HNSW + FTS5)
Concurrent agents	100+ (file-locking)
Memory capacity	Tested 100k+ memories/agent
Batch operations	Atomic, all‑or‑nothing

Common Use Cases

Codebase analysis: retain architecture insights across sessions
Security research: share vulnerabilities across agents
Long‑running research: accumulate findings over days/weeks
Compliance: audit trail of agent decisions
Offline environments: air‑gapped or regulated teams

Quick Start

afs init
afs memory create --agent-id researcher --content "Found SQL injection" --type observation
afs query search --agent-id researcher --query "security"
afs memory share --agent-id researcher --memory-id <id> --swarm-id security-team

Next Steps

Read Architecture to understand internals
Explore the Core Concepts guide for memory, query, graph, and session flows
Use the CLI Reference for complete command coverage