Hindsight Model Card
这是可索引项目报告证据页：它保留 Hindsight Model Card 的源材料入口、机制线索和限制提醒；正文仍需 reader/editor 与 academic public-copy review 后才能当作最终结论引用。
Hindsight Model Card
| Field | Value |
|---|---|
| Repository | vectorize-io/hindsight |
| Category | Learning Agent Memory System |
| Stars / forks snapshot | 14400 / 821 |
| Language | Python |
| License | MIT |
| Raw capture | raw-github/vectorize-io_hindsight.md |
| Updated by | hourly public metadata update, 2026-05-25 11:20 +0800 |
1. Role in Self Evolve
Hindsight is an agent memory system focused on agents that learn over time, with retain/recall/reflect APIs, memory banks, hybrid retrieval and LongMemEval-style performance claims.
2. Working Principle
retain -> fact/entity/time extraction -> hybrid semantic/keyword/graph/temporal recall -> reflect -> learned mental models
3. Evidence Path
web GitHub page observed 1,408 commits, MIT license, Python/TypeScript/Rust stack, 14.4k stars and 821 forks, v0.6.2 latest on 2026-05-14; README describes retain/recall/reflect, memory banks, semantic/keyword/graph/temporal retrieval, LongMemEval claims and coding-agent documentation skill support. Shell GitHub API access remained DNS-blocked and the local gh token was invalid in this run, so this card treats the current snapshot as web-observed rather than API-verified.
4. Teaching Use
Use this card to explain how Hindsight fits the raw -> classification -> project card -> site/report pipeline. It is useful for comparing whether self-evolution is implemented as memory substrate, trace learning, context graph grounding, or benchmark/evaluation infrastructure.
5. Limits
当前未克隆源码，未运行 benchmark、skill install flows、memory experiments、MCP servers、agent evolution loops、security scanners 或 production deployments；star/fork/commit/release 快照来自公开 GitHub 页面文本或可见页面片段。