Agent Swarm Model Card
这是可索引项目报告证据页：它保留 Agent Swarm 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。
Agent Swarm Model Card
One Sentence
Agent Swarm is the clearest production-oriented swarm runtime in the current packet: lead-worker delegation, Docker isolation, MCP, memory, and recurring workflows in one public system.
Three Sentences
It matters because the user explicitly asked for an agent-swarm / swarm-evolve lane, not only single-agent loops. The public repository still exposes a compounding lead-worker runtime where identity, memory, workflows, and external channels all become reusable swarm infrastructure. The 2026-06-17 morning packet keeps the swarm evidence chain aligned to newer stars, fewer open PRs, and another commit on the public runtime surface.
Model Card
| Field | Value |
|---|---|
| Repository | desplega-ai/agent-swarm |
| Source | raw-github/desplega-ai_agent-swarm.md |
| Category | Compounding Lead-Worker Agent Runtime |
| Pattern | ingest tasks from external channels -> lead agent plans and delegates -> workers run inside isolated Docker environments -> shared memory and identity accumulate across sessions -> pages, PRs, replies, and scheduled workflows turn learnings into reusable operations |
| Evidence | Authenticated GitHub API snapshot, 2026-06-17 08:29 +0800 |
Teaching Use
Use Agent Swarm to explain the runtime side of swarm evolution. Improvement is not just inside a model; it also lives in delegation topology, worker isolation, shared memory, workflow packaging, and operational feedback channels.
Evidence And Limits
The raw capture now reflects a GitHub metadata packet observed on 2026-07-05: 573 stars, 65 forks, 1,756 commits, 0 open issues, and 3 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.