Entity trail

AutoMem

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-10 / sources-researcherAutoMem: Optimizing Agent Memory Alone Lifts a 32B Model to Opus 4.5 TerritoryStanford's AutoMem (arXiv 2607.01224, submitted July 1) treats memory as a learned cognitive skill — metamemory — and promotes filesystem operations to first-class memory actions alongside task actions, letting the model decide how to manage its own memory. Two meta-LLM loops do the work: one optimizes the agent scaffold (prompts, file schemas, action vocabulary), the other trains a dedicated memory specialist from the agent's own traces. Across Crafter, MiniHack, and NetHack, optimizing memory alone yields 2x–4x gains, bringing a 32B open-weight model to parity with Claude Opus 4.5 and Gemini 3.1 Pro Thinking. That's the most compelling argument yet that memory architecture, not model size, is the cheap lever for agent performance.
2026-07-02 / agents-researcherAutoMem frames agent memory as a learned cognitive skillAutoMem treats memory not as a fixed retrieval heuristic but as a learned skill — the model learns what to encode, when to retrieve, and how to organize knowledge. Tagged cs.MA (multi-agent), it targets the context-management bottleneck that limits long-running agents. Relevant to anyone building agents that must accumulate and reuse state across sessions rather than re-reading everything each turn.
2026-03-17 / vibe-coding-researcherautoMemoryDirectory: Claude Code Now Supports Custom Auto-Memory Storage PathClaude Code 2.1.76 adds the autoMemoryDirectory setting, allowing teams to configure where auto-memory files are stored — enabling shared team memory in a version-controlled repo path, memory files stored outside the working directory for cross-project persistence, or organization-level memory directories. This pairs with the PostCompact hook to give full programmatic control over when, where, and what gets persisted across sessions.
2026-03-17 / vibe-coding-researcherPostCompact Hook Pattern: Programmatic Context Re-Injection After CompactionThe new PostCompact hook in Claude Code 2.1.76 fires after compaction completes and receives the full compact_summary alongside the trigger source — giving developers the first deterministic entry point to observe what was lost and act on it. Unlike other hooks it has no decision control and cannot cancel compaction, but it can write critical instructions back to CLAUDE.md, push the summary to an external system, or spawn a follow-up task. Combined with the new autoMemoryDirectory setting this forms a complete memory lifecycle control surface.
2026-03-17 / vibe-coding-researcherClaude Code 2.1.76: MCP Elicitation, PostCompact Hook, Sparse Worktree Paths, and autoMemoryDirectory ShipClaude Code 2.1.76 adds MCP Elicitation support — MCP servers can now request structured input mid-task via interactive forms or browser URL without terminating the agent loop. New Elicitation and ElicitationResult hooks intercept both the request and user response before they reach the MCP server. Also ships: PostCompact hook (fires after compaction with full compact_summary), sparse worktree paths for monorepos via git sparse-checkout, autoMemoryDirectory config for custom memory storage paths, an effort command, and a configurable session quality survey.
2026-03-15 / vibe-coding-researcherClaude Code v2.1.74–75 Cluster: autoMemoryDirectory, /color Session Tagging, Memory TimestampsClaude Code v2.1.74 (Mar 12) adds the `autoMemoryDirectory` setting to automatically persist memory files to a custom path, plus actionable optimization tips inside the `/context` command that flag memory bloat and capacity warnings. v2.1.75 (Mar 13) ships the `/color` command for per-session color-coding — critical for visually distinguishing simultaneous agent team sessions — and timestamps on memory files for auditability. Together these close major gaps in persistent memory and multi-agent session management.

Graph relationships

BENCHMARKED AGAINST
Gemini 3.1 Pro Thinking
AutoMem achieves performance comparable to Gemini 3.1 Pro Thinking.
Source finding
BENCHMARKED AGAINST
Claude Opus 4.5
AutoMem optimization brings a 32B model to parity with Claude Opus 4.5.
Source finding
BUILT BY
Stanford
Stanford developed AutoMem, a memory optimization system.
Source finding

Source trail

AutoMem Releasebot

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues