Phases
Roadmap
Slab is built in phases. The first three are done and running on the testnet. We're in the fourth, durability. The order after that is decided: operators first, agents after.
- Phase 1 · done
Foundations
The data plane on its own.
- Append-only segment store with an index, Merkle accumulator for audits
- Client-side encryption (ChaCha20-Poly1305) and Reed-Solomon erasure coding
- QUIC transport, Kademlia routing, hedged reads (first verified shards win)
- Phase 2 · done
Paying for storage without watching it
A control plane that handles money and never sees content.
- Blind-signed vouchers (RFC 9474), checked offline by nodes
- Epoch audits with Merkle proofs, payouts for storage and for delivery receipts
- Repair under churn, proof of possession, stake and slashing
- Phase 3 · done
Testnet
Running it for real, and measuring it.
- Container image and a 20-node testnet on two hosts
- Directories, inline small files, range reads, feeds, a Bee-compatible HTTP gateway
- Measured against Swarm's Bee on CPU, memory, code size and latency
- Self-updating nodes from signed releases, with automatic rollback
- Phase 4 · now
Durability
Making data loss a non-event, at any network size.
- doneOne shard per node, at most 6 shards of a slab per host
- doneJoining through bootstrap.slab.me.uk, saved peers, reachability checks before a node takes shards
- donePlacement hardening from the security review
- in progressParallel slab reads and writes: built, rolling out (a 1 GiB read goes from about 55 to about 7 minutes in our model)
- nextRS 32+16 with 1 MiB slabs for large files, 10+6 kept for small ones
- nextOperators as failure domains: no operator holds more than its share of a slab
- Phase 5
Nodes at home
A node behind a home router, with no port forwarding.
- Nodes learn their public address from peers
- Hole punching through rendezvous peers
- Opt-in, bandwidth-capped relays for the NATs that can't be punched
- Phase 6
Operator tools
Running many nodes from one place.
- A dashboard per account - reachability, stored data, audits and earnings per node
- Node-only deployments and bulk staking
- Phase 7
Operator economics
A reason to run a node, and a price people will pay.
- Prices finer than whole cents per voucher, so a market price can go as low as operators allow
- Payouts from what was actually paid, a market price set by how full the network is
- Storage credit for operators, and starting without a deposit
- Packages for home servers (Umbrel, CasaOS, TrueNAS) and a static binary
- Phase 8
Agents
Storage an AI agent can use on its own, within limits you set.
- Budget wallets with a spending cap the agent can't raise
- An MCP server - put, get and list files with local handles
- Private feeds and agent memory, with forgetting by destroying a key
Not yet
Real money, a public source release and a production network. The testnet can be reset. Moving real money
needs proper payment handling first, and nothing on that path ships without explicit sign-off.