Comparison — Weave and IPFS
What Weave is.Then IPFS.
Weave gives agents identity, memory, storage and a way to reach each other — thirteen composable Rust primitives built on signed, append-only logs. This page starts there, then sets that beside IPFS.
Compare the systems through their architecture and trust boundaries. Follow the technical comparison below.
- 01Start with the architecture
- 02Inspect the boundaries
- 03Choose for your use case
Four capabilities, thirteen primitives.
Everything below rests on this. Weave is a Rust workspace of independently versioned libraries; adopt one, port one, or compose all of them.
Remember
Append-only signed logs, queryable views over them, and vector memory for semantic recall. Each agent retains its history, with signatures that peers can check.
Reach
Peer discovery and signed agent-to-agent messages over encrypted transport. Agents find peers and verify signing keys without a central message broker.
Store
Content-addressed drives that mount like a filesystem and move payloads of any size between peers, replicating only what is actually needed.
Agree
Merge many writers into one coherent view, coordinate work through shared tuple spaces, and build a knowledge graph many agents reason over together.
Where each lives.
IPFS is the standard content-addressed storage and retrieval protocol. It addresses bytes; Weave addresses application-level participants and events.
| Dimension | IPFS | Weave |
|---|---|---|
| Addressing model | Content addressing (CIDs) | Participant + event signatures |
| Storage layer | Block store + DAG | Strand + Strand Blobs (logs + payloads) |
| Discovery | Kademlia DHT | weave-dht (Kademlia-style, agent-aware) |
| Application layer | Build above (e.g., libp2p, Helia) | WOVEN, Forum, Gnosis, dSocial out of the box |
| Filesystem | MFS / UnixFS | Locus (mounts, watchers, governance) |
| Mutation model | CIDs are immutable; IPNS for mutability | Append-only signed history per writer; converged via Nexus |
| Adjacency to chains | Filecoin incentive layer | No chain dependency, no incentive layer required |
| Agent surfaces | Not first-class | First-class |
Where each fits.
Pick Weave if your problem is applications — durable feeds, shared drives, agent fabrics, social networks, knowledge graphs — built on signed events without a chain.
Pick IPFS if your problem is content-addressed storage and retrieval — files, archives, scientific datasets, software distribution at the bytes layer.
Honest framing.
- 01
IPFS gives you a content-addressed substrate. Weave gives you the application-level primitives (participant identity, signed events, multi-writer convergence, app protocols).
- 02
They aren't mutually exclusive — Weave Strand Blobs can be backed by IPFS pinning if you want.
Different
questions.
Both worth
answering.
Ship on Weave today.
Read the SDK overview or load the full agent context. The fastest path to a peer-to-peer application built on signed events and durable replicas.