Buzz: An Open‑Source Collaboration Platform Uniting Humans and AI Agents in One Room

Three Core Concepts

Buzz builds on three Nostr concepts:

Nostr – a decentralized communication protocol where each participant is identified by a public‑private key pair and messages are signed.

Signed event – the basic Nostr data unit; any chat message, like, workflow step, code review, or git event is an event signed with the publisher’s private key.

Relay – a server that receives, stores, and forwards events according to subscription filters. Relays do not own the data.

What is Buzz

Buzz is a self‑deployable team‑collaboration space that runs on a self‑owned relay, allowing humans and AI agents to share rooms, message streams, and audit logs.

Every interaction—chat, like, workflow step, code review, git event—is written as a Nostr signed event stored in a single log. Humans and agents use distinct key pairs but share the same event format, retrieval mechanisms, and auditability, making agents first‑class members of a room.

Problems It Solves

Traditional AI‑agent integration follows three flawed patterns:

Granting the agent many permissions (large token sets) – risk of over‑privilege and static permission tables.

Isolating the agent in a sandbox – the agent cannot see team context or history.

Connecting the agent via separate bots – each bot has its own identity, making actions non‑traceable.

Buzz replaces permission switches with identity isolation: an agent’s view is determined solely by the channels it joins, just like a new human teammate.

How It Works

Human and agent clients use the same Nostr protocol; the only difference is the signing key. Relays verify signatures and event format without distinguishing humans from agents. Consequently, searches return mixed human and agent activity, audit trails show who performed each step, and agents can participate in code reviews, CI results, and release notes.

Real‑World Scenarios

Scenario 1 – Incident Memory : At 2 am a developer asks “Has this error occurred before?” An agent pulls six months of history, posts related posts, root‑cause analysis, and the last author of the buggy code. All evidence stays in the channel for later reference.

Scenario 2 – Branch‑as‑Room : Creating a feature branch spawns a channel; patches are posted as NIP‑34 signed events, CI results flow into the channel, the agent performs the first review, and teammates can like or comment. The decision to merge and its evidence stay together.

Scenario 3 – Automated Release Notes : Tagging a commit triggers a workflow; the agent drafts release notes from merged PRs, sends them to a human reviewer, and upon approval automatically publishes them. Every step is signed and traceable.

Technical Architecture

Buzz is a Rust workspace composed of multiple crates. Key design choices:

Rust – zero‑cost abstractions and memory safety enable high‑concurrency WebSocket handling without GC pauses.

Postgres – provides full‑text search, channel filtering, and author queries, avoiding a separate search engine.

Tauri – uses the system WebView for a lightweight desktop client, reducing resource usage compared to Electron.

buzz‑cli JSON I/O – designed for LLM tool calls; input and output are structured JSON rather than mixed‑format CLI text.

Core Capabilities (Collaboration Layers)

Communication Layer – channels, threads, direct messages, media comments (including frame‑level video comments).

Identity Layer – humans and agents each have a key pair; joining a channel follows the same process for both.

Audit Layer – all events are signed, stored in a hash‑chain audit log, detectable even if the main database is tampered.

Automation Layer – YAML‑defined workflows triggered by messages, reactions, schedules, or webhooks; each step is a signed event.

Git Integration Layer – uses NIP‑34 to represent patches, repo announcements, and status updates as events; branches become channels.

Agent Integration Layer – supports Goose, Codex, Claude Code, etc., plus a shell/file editing tool and persona configuration.

Installation & Getting Started

Pre‑built binaries are available on the GitHub Releases page for macOS, Linux, and Windows. The default relay address is ws://localhost:3000. To point to a custom relay, set the BUZZ_RELAY_URL environment variable or switch inside the app.

git clone https://github.com/block/buzz.git && cd buzz
just setup && just build
just dev   # launches relay and desktop client

Agents authenticate by providing a BUZZ_PRIVATE_KEY and interacting via buzz-cli.

Advantages

Unified identity model: one key pair, one event log, one search index for chat, code, CI, and releases.

Native auditability: signed events form a tamper‑evident hash‑chain.

Agents are first‑class members, not external bots.

Self‑hosted data ownership.

Active Rust implementation: frequent commits (latest on 2026‑08‑02) and rapid releases (v0.5.3 on 2026‑07‑31).

Limitations & Risks

Feature gaps: mobile clients, full workflow approval glue, and push notifications are still in progress.

Build complexity: requires Rust, Docker, and Node tooling.

Open issues: 1,544 open issues indicate active but immature development.

Ecosystem: only four workflow triggers and limited third‑party integrations compared to Slack’s extensive app directory.

Comparison with Similar Tools

Agent Identity – Buzz: formal member with key and audit trail; Slack/Discord/IRC bots: external programs without unified identity.

Audit – Buzz: signed events + hash‑chain, tamper‑proof; Slack/Discord: server logs requiring trust.

Data Ownership – Buzz: self‑hosted, fully owned; Slack/Discord: SaaS‑hosted data.

Git Integration – Buzz: NIP‑34 patches/reviews/CI in channel; Slack/Discord bots rely on third‑party GitHub apps.

Search – Buzz: full‑text search across human and agent events; Slack/Discord: message‑only search, bot actions not searchable.

Workflow – Buzz: YAML, event‑triggered, steps signed; Slack: Workflow Builder; Discord and IRC: no native workflow.

Deployment – Buzz: self‑hosted relay; Slack/Discord: SaaS.

Maturity – Buzz: v0.5.x (early); Slack/Discord: mature.

License – Buzz: Apache‑2.0; Slack/Discord: closed source; IRC bots: open source.

Design Takeaways

Identity isolation (channel membership) is safer and more natural than permission tables.

A single unified protocol reduces glue code when integrating multiple systems.

Audit should be an architectural by‑product, not an afterthought.

Separate tooling for humans (UI) and agents (JSON API) improves usability.

Explicitly marking feature status (Works today / Being wired up / Strong opinions) builds trust.

Conclusion

Buzz redesigns collaboration infrastructure by leveraging Nostr’s signed‑event model to place humans and AI agents in the same identity system, event log, and search index. It is not a Slack replacement yet; the project is early‑stage with many open issues and a missing mobile client. Nevertheless, its concepts—identity‑based isolation, protocol‑level unification, and built‑in audit—offer valuable lessons for embedding AI agents into team workflows.