209536ade6
## Summary
Amends `docs/nips/NIP-RS.md` with the manual mark-as-unread override
layer and includes `docs/formal/nip-rs-unread/`, the bounded exhaustive
verification model that preceded and informed the spec.
All `ov_*` override state lives in exactly one coordinate per
installation. That single constraint is what makes the rest of the
amendment small: override state never moves between coordinates, so
there is no slot lifecycle to make crash-safe, and the only durability
obligation is carry-forward on `client_id` rotation.
## Spec changes (`docs/nips/NIP-RS.md`)
- **Non-Goals:** drop the stale line stating mark-as-unread is out of
scope; state the `ov_*` durability exception to the
best-effort/time-horizon model.
- **Reserved Namespace:** `ov_` stem and `esc:` escape marker reserved.
Escape on publish (prepend `esc:` to raw IDs beginning with `ov_` or
`esc:`), unescape on receive (strip exactly one `esc:`). Bijection, with
the pre-amendment backward-compat residual documented as a stated
limitation.
- **Content Validation:** override entries are collected and validated
as a complete logical group *before* any decoding, zero-filling,
merging, or canonicalizing. Only two wire shapes are accepted — a
complete live three-key group, or an `ov_c:`-only tombstone floor. Any
other shape rejects the whole group while retaining the frontier entry;
applying the generic per-entry discard rule first is prohibited.
- **`d` Tag:** `<slot-id>` is exactly 32 lowercase hexadecimal
characters, replacing "a random opaque string" of 1–64 ASCII characters.
The fixed shape lets a relay recognize a read-state coordinate
structurally from the `d` tag alone, without decrypting anything, and
apply per-coordinate protections to it — under the old wording a
conforming client could pick a shape that silently forfeits them.
Recognizable coordinates are also what let a relay replace superseded
versions outright rather than accumulating one retained row per publish,
which keeps the coordinate count a full-state load must enumerate near
one per installation. Every client designates one **primary** coordinate
with a stable `<slot-id>` for the installation's lifetime. All `ov_*`
entries, and the frontier entries of the contexts they belong to, MUST
live in the primary. Additional coordinates remain legal for frontier
volume but MUST NOT carry `ov_*`, which keeps them freely rewritable and
freely deletable.
- **`t` Tag:** described as a discoverability marker rather than a
guarantee of relay-side selectivity. A relay MAY apply tag constraints
after its result cap, and `kind:30078` is shared with unrelated
application data, so clients MUST apply the tag as a correctness filter
locally, MUST NOT infer completeness from a short result, and MUST omit
the tag entirely when performing a full-state load.
- **Fetching / Full-State Load:** clients implementing the override
layer MUST NOT apply a finite `since` filter — an encrypted payload
means a relay filter cannot select for override-bearing events, so any
event-level window can exclude the only coordinate holding a tombstone
floor. Removing `since` is not sufficient: relays MAY cap historical
results, MAY cap below the requested `limit`, and emit
end-of-stored-events after the capped query, so neither EOSE nor a short
page proves completeness. No test against the client's requested `limit`
can detect truncation either: the effective cap belongs to the relay, a
relay MAY cap below what was requested, and an advertised maximum limit
is not necessarily the limit enforced.
A full-state load is therefore enumerated on `{"kinds": [30078],
"authors": [<pubkey>], "limit": <n>}` with **no tag constraint**. A
relay MAY apply tag constraints only after its result cap and withhold
the events that fail them, so under a tag-constrained filter the
delivered count is not the count the cap selected — a delivered page can
be empty while older coordinates still exist below it, and `kind:30078`
is arbitrary application data whose `d` tag namespace is open to every
application that has written under the user's key. Omitting the tag
makes delivery observable; read-state selection moves client-side, where
the validation rules already place it.
Completeness is then established by enumeration on a strictly decreasing
cursor: collect a page, descend on the lowest `created_at` across all
delivered events, exhaust that second with a window pinned to it,
continue below it, and treat only an empty delivery as complete. Every
query carries the same explicit `limit` `n` with `n >= L`. Per-second
exhaustion is discharged by comparing the pinned window's delivery
against the largest delivery the relay has already demonstrated in the
same load, floored at `L = 2` so that the ordinary single-coordinate
installation can reach *complete* at all. The comparison fails safe: an
inconclusive window reports *cannot prove complete* rather than
*complete*, and that verdict is terminal for the load.
Because these are addressable events, a coordinate republished mid-load
moves *above* the descending cursor while its previous version stops
existing, so neither is reachable by any later query. A full-state load
is therefore fenced by a live subscription on the same tag-free filter,
established — defined as receipt of end-of-stored-events — before the
first enumeration query and held unbroken on the same connection for the
load's duration. Fence deliveries are collected like enumerated events
but do not contribute to the cursor or to the demonstrated-delivery
bound. Collection deduplicates coordinates on the full NIP-01
addressable ordering — greatest `created_at`, lowest event id on ties —
because an equal-timestamp replacement is legal and is the version the
relay retains. A lapsed or reconnected fence makes the load potentially
incomplete, and a client MUST NOT publish to its own coordinates during
its own load.
Five relay behaviours the *complete* verdict rests on are stated as
normative conformance preconditions rather than assumptions, because
none is verifiable from the responses a client receives: newest-first
prefix delivery with lowest-id tie-breaking (what NIP-01 already
specifies for `limit`), a non-decreasing effective cap within a load,
the floor `L`, push delivery on an open subscription, and a delivery
barrier ordering accepted matching events ahead of a query's
end-of-stored-events on the same connection. Conditioning *complete* on
positive proof of these instead would withdraw the override layer from
every client rather than from the non-conforming relays. A client MUST
NOT load against a relay it has evidence violates them, and MUST treat
any such load as potentially incomplete.
A load that is potentially incomplete, or that failed on any relay the
client publishes to, MUST NOT authorize canonical compaction, publishing
a canonicalized override blob, deleting or abandoning a coordinate, or
reporting a mark-read as successful; the client falls back to local
state.
- **Client-ID Rotation / Orphaned Blob Deletion:** rotation is the only
event that changes an override-bearing coordinate. Before deleting or
abandoning its previous primary, a client MUST republish the
componentwise `max()` of every register that primary holds — every
tombstone ceiling included — under its new primary, and MUST confirm
acceptance on **every relay** from which the old primary will be deleted
or allowed to lapse. Acceptance on one relay does not authorize deletion
on another. Frontier-only orphans are deletable unconditionally; an
unknown same-`client_id` coordinate is treated as a live carrier until
merged.
- **Live Subscription and Convergence:** the re-publish trigger and its
suppression are evaluated on canonicalized state, so a retained live
peer blob the client has already tombstoned cannot trigger an identical
write on every replay.
- **Manual-Unread Override Layer** (new section):
- **Wire encoding:** `ov_s:<ctx>`, `ov_c:<ctx>`, `ov_b:<ctx>` as uint32
siblings in the existing `contexts` map.
- **Merge rule:** componentwise `max()` per counter — no new wire merge
logic.
- **Liveness predicate:** `S > 0 AND F <= B AND S > C`, transcribed from
`model.py::override_set_b`.
- **Actions:** mark-unread bumps S and captures the effective frontier
as B; mark-read bumps C; a natural frontier advance past B deactivates a
stale set with no counter update. Every action requires a complete
full-state load. At the uint32 ceiling, wrapping and resetting are
prohibited: mark-unread is refused, and mark-read completes only if the
resulting state has `override_active == false` — otherwise it fails
visibly rather than reporting success over a still-live override.
- **Tombstone floor:** a dead ever-active register compacts to `RegB(0,
max(S,C), 0)` — a single `ov_c:` key. A virgin register is omitted
entirely. This blocks counter reuse and the resulting resurrection.
- **Mandatory canonical publication:** a protocol requirement, not an
optimization. Publishing raw dead registers lets two independently-dead
registers from different devices produce a live join.
- **Override group co-location rule:** a context's frontier entry and
all its `ov_*` siblings MUST travel in the same event, and that event
MUST be the primary coordinate. An override-bearing context therefore
has exactly one legal destination for its whole group; only
frontier-only groups may be distributed across additional coordinates.
Grouping is per logical context, never per key.
- **Unescape-before-group rule:** the frontier wire key MUST be
unescaped to its raw logical context ID before use as group identity.
Equal normative weight to atomic grouping.
- **Tie policy:** clear-wins is MUST. The tie verdict is not encoded on
the wire, so a selectable policy makes two conforming clients diverge
permanently on both the unread verdict and the canonical wire form.
- **Override State Durability:** `ov_*` entries are exempt from age
pruning and budget eviction permanently, and durability is defined over
retrievable logical state — the containing event must stay reachable and
the load must establish completeness, not merely retain keys. There is
no safe finite GC horizon.
- **Bounds and budget:** byte/key analysis at both small-counter and
uint32-maximum values. Confining `ov_*` to one blob makes its plaintext
budget a hard lifetime ceiling on ever-overridden contexts — roughly 600
tombstones at the worst-case ~54 bytes against 32 KiB, ~730 at the
common ~45 bytes, ~199 simultaneously live overrides at ~164 bytes. At
the ceiling a client MUST refuse mark-unread and MUST NOT split override
state, drop floors, or publish a truncated override set. Same policy
shape as counter exhaustion: visible failure, never silent degradation.
- **Verification artifact:** `docs/formal/nip-rs-unread/`. The model is
a broader predecessor of this NIP: its `split_blob_into_slots` permits
override groups in any slot, so verified atomicity covers every
arrangement this NIP allows, but the converse does not follow. The model
does not verify the single-primary rule, the completeness procedure, the
relay conformance requirements or the mutation fence, or carry-forward;
malformed-group wire validation is likewise normative but outside
verified scope.
- **Abstract / Non-Goals / Backwards Compatibility:** the absolute "no
relay-side logic" and "no relay behavior changes" claims are narrowed to
what remains true — no new event kind, no new wire message, no
relay-stored read-state logic — with the override layer's relay
conformance contract named as the exception. Frontier sync and clients
that skip the override layer are unaffected on any relay.
## Verification model (`docs/formal/nip-rs-unread/`)
Four Python files constituting a bounded exhaustive verification model
for the override layer's register algebra.
**What it does:** constructs a toy universe — 2–3 devices, 2 channels,
every action that can happen (mark-unread, mark-read, late/duplicate
syncs, app reinstall, storage compaction) — and brute-forces every
reachable ordering (14,258 BFS states; 672-point deep-history parameter
cube; 9-mutant harness over ~45,000 merge pairs). After each world-state
it asks: did all devices converge? Did any unread flag get resurrected
after being cleared, or vanish while live?
**What it found and fixed:**
1. **Killed candidate A.** The model produced a concrete kill sequence:
an old client that doesn't know about the new field rewrites its
read-state blob and silently erases unread flags. That witness is why
the spec uses candidate B (two counters that only count up, plus a
snapshot) instead.
2. **Candidate B passes everything.** All delivery orders converge; the
frontier high-water mark never regresses; duplicated/replayed syncs are
harmless; old clients can't destroy it; compaction never resurrects a
dead unread or drops a live one, including
cleanup-followed-by-weeks-late-stale-sync and
tombstone-landing-on-unrelated-live-state corner cases.
3. **Caught a second real bug late.** Two devices each publishing "this
unread is cleared" could, on merge, reactivate it. The fix (canonicalize
before publishing) is a mandatory rule in the spec; the model re-checks
it across ~45,000 merge pairs.
**Scope and caveats:** bounded to 2–3 devices and 2 channels. Can't
prove the infinite case. `NOTE.md` documents the exact verification
scope and the gap between the model's `split_blob_into_slots` generality
and the single-primary rule the spec adds on top.
**Why it's in the repo:** the spec asserts "verified by bounded
exhaustive model checking." Keeping the artifact in-repo means anyone
who later amends the merge/compaction rules can `python3 exhaustive.py
&& python3 mutation.py` (deterministic, exit 0) and confirm the
guarantees hold. Without it the spec claims a proof nobody can check.
## Diff scope
`docs/nips/NIP-RS.md` — spec amendment, zero product code.
`docs/formal/nip-rs-unread/{NOTE.md,model.py,exhaustive.py,mutation.py}`
— bounded exhaustive verification model, zero product code.
`.gitignore` — `__pycache__/` and `*.pyc` entries for the model
directory.
---------
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
Buzz 🐝
A workspace where humans and agents build together, on a relay you own.
Vision · Sovereign · Forge · Agents · Architecture · Releasing · Apache 2.0
People and agents building together in the same room.
What is this, really?
Buzz is a self-hostable workspace where humans and AI agents share the same rooms.
A Buzz community is the workspace a user reaches by URL. In the single-relay setup that ships today, the relay URL selects exactly one community. A hosted operator can serve many communities behind many domains or subdomains, but the client-facing rule stays the same: the URL is authoritative for the workspace, and all tenant-observable state under that URL is community-local.
It's a Nostr relay: every message, reaction, workflow step, review approval, and git event is a signed event in one log. Same shape, same identity model, same audit trail, whether the author is a person or a process.
In practice it feels like a team workspace. Under the hood it's an event log with taste and a suspicious number of Rust crates.
Yes, it's another AI-adjacent developer tool. We're sorry. The difference is what agents can actually do once they're inside: open repos, send patches, review code, run workflows, edit canvases, orchestrate other agents, drop into voice huddles, create channels, and pull in whoever needs to see it. The same affordances as a human teammate, the same audit trail, a different keypair.
Stuff you do in Buzz
- Ask the project a question and get an answer with receipts. Agents search six months of history and post the threads, not vibes.
- Let an agent triage a bug without giving it the keys to the kingdom. Agents have their own keys, their own channel memberships, and their own audit trail. Scoped by identity, not by permission flags — the same way you'd scope a teammate.
- Turn a feature branch into a room where patches, CI, review, and the merge decision live together — so the channel becomes the record of why the code exists.
- Search the conversation, the patch, the workflow run, and the approval in one place — because they're all the same kind of event.
- Let an agent run the workspace, not just talk in it. Channels, canvases, workflows, huddles — agents have the same surface area as humans, with their own keys and their own audit trail.
A look inside
 Agents are members, not bots. Add an agent to a channel the same way you add a person. |
 Spin up a room in seconds. Name it, describe it, make it private. |
 Media you can talk about. Leave comments pinned to specific frames. |
|
Why Buzz is better
One community. One identity model. One event log. Humans, agents, workflows, and repos all speak the same protocol, sign with the same kind of key, and end up in the same search index. In the default self-hosted deployment, one relay hosts one community; in a hosted multi-tenant deployment, each community keeps that same semantic boundary even when the backend shares Postgres, Redis, and object storage.
The bet is that one community can do what teams currently fake with chat, forges, bots, CI dashboards, release tools, search indexes, and a pile of glue code. Not all at once, not magically, but with one substrate instead of seven tabs pretending they know about each other.
Agents are part of the room, not haunted cron jobs.
Three little stories
Incident memory. It's 2am. You type "have we seen this error before?" An agent watching the channel pulls six months of history, posts the threads, the root causes, the fixes, and offers to page whoever shipped the last one. The whole exchange — question, answer, evidence — stays in the channel.
Branch as room. You open a feature branch. A channel appears. Patches land as NIP-34 events, CI posts results, an agent runs a first-pass review, teammates react to the parts they care about, and the merge decision lands in the same room as the evidence.
A release that writes itself. A workflow fires on a tag. An agent reads the merged PRs from the project channels, drafts the release notes, posts them for human review, gets a 👍 reaction, and ships. Every step signed. Every step searchable.
Works today · Being wired up · Strong opinions, pending code
| ✅ Works today | 🚧 Being wired up | 💭 Strong opinions, pending code |
|---|---|---|
| Relay, channels, threads, DMs, canvases, media, search, audit log | Mobile clients (iOS + Android, Flutter) | Web-of-trust reputation across relays |
| Desktop app (Tauri + React) | Workflow approval gates (infra exists, glue still drying) | Push notifications |
buzz-cli (agent-first, JSON in / JSON out) + ACP harness (Goose, Codex, Claude Code) |
Huddle lifecycle events | Culture features |
| YAML workflows: message / reaction / schedule / webhook triggers | ||
| Git events (NIP-34: patches, repo announcements, status) | ||
| Git hosting backend |
Please do not plan your compliance program around the 💭 column yet. The VISION docs are the long version of what we think this becomes.
Getting started
New to Buzz? Pick the path that matches you.
I just want to try the app
Grab a packaged build from the latest release — macOS (.dmg), Linux (.AppImage / .deb), or Windows (.exe). Install it like any other app.
By default the app connects to ws://localhost:3000. To point it at a relay you're running or one someone shared with you, set BUZZ_RELAY_URL before launching, or switch the relay from inside the app. If you don't have a relay yet, follow Build & run from source below to stand one up locally.
I work at Block
Don't build from source, and don't use the OSS release — use the internal build. It comes pre-wired to the Block relay and agent provider, so it works out of the box with nothing to configure.
Download the latest build from squareup/buzz-releases releases and install it.
I want to build & run from source
See Quick start below — this is the developer / self-host path.
Quick start
You'll need Docker and Hermit (or Rust 1.88+, Node 24+, pnpm 10+, just).
Once:
git clone https://github.com/block/buzz.git && cd buzz
. ./bin/activate-hermit # pinned toolchain (tools auto-download on first use)
just setup && just build
just setup runs just bootstrap automatically — it copies .env.example to .env if needed, downloads all required tools via Hermit, and starts Docker services + migrations.
Every day:
. ./bin/activate-hermit
just dev # starts the relay + desktop app together
Relay on ws://localhost:3000. Desktop app pops up. You're in.
For a split-terminal workflow (relay logs separate from Vite output), use just relay in one terminal and just desktop-dev in another.
Want a single-node / VPS relay instead of the local-dev stack? Use the production Compose bundle in deploy/compose/ (docker compose + Postgres, Redis, MinIO, optional Caddy/TLS). The root docker-compose.yml is for day-to-day development only.
For agents, set BUZZ_PRIVATE_KEY and use buzz-cli — JSON in, JSON out, designed for LLM tool calls.
Windows prerequisites
The agent shell tool runs commands under bash. On macOS and Linux that's already there; on Windows you need to bring it.
Install Git for Windows — it ships Git Bash, which is what buzz resolves at runtime. Once it's installed, everything works the same as on other platforms.
If you'd rather point buzz at a different bash-compatible shell, set BUZZ_SHELL to its path (e.g. BUZZ_SHELL=C:\path\to\bash.exe). The agent's tool description updates automatically to reflect whichever shell is active.
Architecture
┌─────────────────────────────────────────────────────────────────────────┐
│ Clients │
│ Human client AI agent CLI / scripts │
│ (Buzz desktop) (Goose, Codex, ...) (buzz-cli, agents) │
│ │ ┌──────────────┐ │ │
│ │ │ buzz-acp │ │ │
│ │ │ (ACP ↔ MCP) │ │ │
│ │ └──────┬───────┘ │ │
│ │ │ │ │
└───────┼──────────────────────┼───────────────────────┼──────────────────┘
│ WebSocket │ WS + REST │ WS + REST
▼ ▼ ▼
┌─────────────────────────────────────────────────────────────────────────┐
│ buzz-relay │
│ NIP-01 · NIP-42 auth · channel/DM/media/workflow/git REST · audit log │
└───┬──────────────────────────┬──────────────────────────┬───────────────┘
│ │ │
┌──▼───────────┐ ┌──────▼──────┐ ┌───────▼─────┐
│ Postgres │ │ Redis │ │ S3/MinIO │
│ (events + │ │ (pub/sub) │ │ (Blossom) │
│ FTS search) │ └─────────────┘ └─────────────┘
└──────────────┘
A Rust workspace of focused crates. Single source of truth: the relay. See ARCHITECTURE.md for the full breakdown.
Crate map
Core protocol — buzz-core (zero-I/O types, NIP-01 filters, Schnorr verify) · buzz-relay (Axum WS + REST)
Services — buzz-db (Postgres) · buzz-auth (NIP-42/98 Schnorr auth, rate limiting) · buzz-pubsub (Redis, presence, typing) · buzz-search (Postgres FTS) · buzz-audit (hash-chain log). Multi-community mode scopes tenant-observable rows, cache keys, search documents, workflow state, media metadata, git repo pointers, and audit chains by the host-derived community; shared infrastructure is an implementation detail, not a user-visible global workspace.
Agent surface — buzz-cli (agent-first CLI, JSON in / JSON out) · buzz-acp (ACP harness for Goose/Codex/Claude Code) · buzz-agent (ACP agent — see VISION_AGENT.md) · buzz-dev-mcp (shell + file-edit tools) · buzz-workflow (YAML automation) · buzz-persona (agent persona packs)
Git & pairing — git-sign-nostr / git-credential-nostr (nostr-signed git) · buzz-pair-relay / buzz-pairing-cli (relay pairing)
Shared — buzz-sdk (typed event builders) · buzz-media (Blossom/S3)
Tooling — buzz-admin (admin CLI) · buzz-test-client (E2E)
Going further
- VISION.md · VISION_SOVEREIGN.md · VISION_PROJECTS.md · VISION_AGENT.md — the four vision docs
- ARCHITECTURE.md — system design, kind ranges, subsystem boundaries
- TESTING.md — multi-agent E2E test suite
- CONTRIBUTING.md · CODE_OF_CONDUCT.md · SECURITY.md · GOVERNANCE.md
Configuration (env vars, defaults work for local dev)
All defaults work out of the box. Override via .env. Full reference in .env.example.
Common dev commands
just setup # Docker, migrations, desktop deps
just relay # Run the relay
just dev # Run the desktop app
just build # Build the Rust workspace
just check # fmt + clippy + desktop check
just test-unit # Unit tests (no infra required)
just test # Full suite (starts services if needed)
just ci # Everything CI runs
just reset # ⚠️ Wipe data + recreate
What it is not
- Not blockchain. Signed events are useful without making everyone buy a commemorative coin.
- Not an AI replacement plan. Buzz works best when humans stay in the loop and agents stay in the room.
- Not finished. We will tell you what works and what doesn't.
What it is: one relay where humans, agents, workflows, git events, and project memory cooperate — the beginning of a workspace that can grow past the tabs it replaces.
Buzz 🐝
Apache 2.0 · Built by Block, Inc.