Head to head

ArgosBrain vs Claude Code's CLAUDE.md

CLAUDE.md lives in the system prompt. ArgosBrain lives in retrieval.

What Claude Code's CLAUDE.md does

Tiered MD files loaded into the system prompt at session start.

What ArgosBrain does differently

Retrieval-time graph — pays when queried, not every turn, with symbol precision and auto-invalidation.

What Claude Code's CLAUDE.md is

The baseline, stated fairly.

Claude Code (Anthropic) ships tiered memory: enterprise policy → project ./CLAUDE.md → parent directory CLAUDE.mds → user ~/.claude/CLAUDE.md → per-project auto-memory at ~/.claude/projects/<hash>/memory/MEMORY.md. Since Feb 2025 for CLAUDE.md; auto-memory layer in the 2.x series (late 2025).

Imports via @path/to/file.md, up to 5 levels deep. Slash command /memory, quick-add with # prefix.

How it actually works

Technical facts.

StorageMD files on disk.
RetrievalLoaded into system prompt on session start — not retrieved per query.
CostPrompt tokens on every single turn, equal to the size of the loaded memory.
LimitationsNo symbol awareness, no decay, no contradictions, no cross-project graph without symlinks.
Copied byCursor (.cursorrules → AGENTS.md), Zed (.rules), Aider, Cline, Continue, OpenAI Codex CLI (AGENTS.md). The agents.md spec generalizes this.

Sources: Claude Code memory docs

Verdict

Where each one wins.

↑ Where ArgosBrain wins

Retrieval vs system prompt. CLAUDE.md pays every turn. ArgosBrain pays when queried.
Symbol precision. CLAUDE.md is prose.
Auto-invalidation. CLAUDE.md rots until a human edits.

↑ Where Claude Code's CLAUDE.md wins

Zero install — it's Claude Code itself.
Human-readable, git-trackable — you can diff memory.
Great for project conventions, team rules, style guides — stuff that should live in the system prompt.

When to choose which

Honest recommendation.

Choose Claude Code's CLAUDE.md if

Small projects
Conventions, style, team rules

Choose ArgosBrain if

Both: conventions in CLAUDE.md, ingested code structure in ArgosBrain. Complementary, not competing.