Comparison
How Swarm 357 compares to other frameworks.
How Swarm 357 compares
This is a concise, honest positioning reference. For the canonical “Why not X?” table, see README.md.
| Dimension | Swarm 357 | Typical notes |
|---|---|---|
| Memory | Portable Memvid .mv2 via Rust bridge + flat-file .swarm/ | Many frameworks assume a hosted vector DB only. |
| Ontology | Six business layers + management meta-agents | Crew-style crews are often ad-hoc YAML without domain structure. |
| Cost | Per-agent budgets, layer limits, downgrade signals | Not all SDKs ship layered accounting out of the box. |
| Security | BashSecurityGate for shell tooling | Agent frameworks vary; we centralize bash policy. |
| Scale claim | 357 identities from roster + soul templates (bounded concurrency) | Not the same as 357 simultaneous Opus sessions. See README. |
When to pick something else: You need durable LangGraph-style checkpointing, huge PyPI ecosystem plugins, or managed guardrails from a specific vendor SDK. Evaluate those tradeoffs explicitly.