OpenClaw Model Card
这是可索引项目报告证据页：它保留 OpenClaw 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。
OpenClaw Model Card
One Sentence
OpenClaw is the product-runtime sample for asking whether self-evolving agents are actually usable by people.
Three Sentences
The raw capture still describes a local-first assistant across many channels, with tools, sessions, skills, and companion apps rather than only a benchmark harness. It is not primarily a paper-code project; it is a user-facing agent operating-system surface. That is exactly why it belongs in the public evidence chain, and the 2026-06-17 morning packet keeps this product-runtime anchor fresh.
Model Card
| Field | Value |
|---|---|
| Repository | openclaw/openclaw |
| Source | raw-github/openclaw_openclaw.md |
| Category | Personal agent product runtime |
| Pattern | local-first assistant runtime -> multi-channel gateway -> sessions/tools/skills -> companion apps and durable operating surface |
| Evidence | Authenticated GitHub API snapshot, 2026-06-17 08:29 +0800 |
Teaching Use
Use OpenClaw to teach the gap between “agent method” and “agent product.” A usable agent needs channels, permissions, memory, tool access, sessions, scheduling, and recovery paths.
Evidence And Limits
The raw capture now reflects a GitHub metadata packet observed on 2026-07-05: 381,730 stars, 80,030 forks, 64,040 commits, 3,604 open issues, and 3,469 open pull requests. This packet is fresher than the previous authenticated packet at 2026-07-05 01:38 +0800 where a delta was observed. This run did not execute the repository locally, validate workflows end to end, or independently rerun benchmark claims. Product, memory, benchmark, and automation claims therefore remain repository-scoped unless separately tested.