> An engineer is debugging a production incident at 2am. They type in the incident channel: "What happened last time we saw this error?"
>
> An agent watching the channel searches six months of incident history and posts the threads, root causes, and fixes — then offers to page the engineer who deployed the last one.
The platform made it possible. The agent made it happen. Sprout is the pipe — event store, search index, subscriptions, delivery — not the brain. Humans and agents bring the intelligence. Sprout gives them a shared space to use it.
---
## Surfaces
| Surface | Model | Default Notifications |
|---------|-------|-----------------------|
| 🏠 **Home** | Personalized feed. What matters to you. | — |
Guests (investors, reporters, partners) get a scoped token with membership in specific channels. Same access model as everyone else. Guests can connect with their own Nostr client (Coracle, nak, Amethyst) through [`sprout-proxy`](NOSTR.md), which translates standard NIP-28 events to Sprout's internal protocol. Two auth paths: pubkey-based guest registration (persistent) or invite tokens (ad-hoc, time-limited).
[Nostr NIP-01](https://github.com/nostr-protocol/nips/blob/master/01.md) on the wire. Every action — a message, a reaction, a workflow step, a profile update — is a cryptographically signed event:
```
id sha256 of canonical bytes
pubkey secp256k1 public key
kind integer (the only switch)
tags structured metadata
content JSON payload
sig Schnorr signature
```
Sprout extends the standard Nostr event format with custom kind numbers for enterprise features.
New message type? New kind integer. Zero breaking changes.
Rust backend, TypeScript/React desktop. The server is a Cargo workspace of focused crates — relay, auth, pub/sub, search, audit, workflow engine, MCP agent interface, and more. The desktop client is a Tauri 2 app with React 19. See [README.md](README.md) for the full crate map.
One model. TLS in transit. At-rest encryption delegated to the storage layer (e.g., Postgres TDE, volume encryption). Server-managed encryption covers every channel, every DM, every event — eDiscovery works on everything. End-to-end encryption (NIP-44) is a future consideration for DMs.
Channel-scoped YAML-as-code automation with conditional logic — the feature Slack paywalled for 5 years. Message triggers, reaction triggers, scheduled runs, webhooks. Every step traced. Agents manage workflows through MCP tools.
Approval gates are partially built: the schema, REST endpoints, MCP tool, and UI all exist. The executor doesn't yet persist the approval token or suspend execution — a run that hits a `request_approval` step is marked Failed (WF-08). The infrastructure is there; the wiring is next.
The Home Feed is the personalized entry point — @mentions, items needing action, channel activity, agent updates. Fan-out-on-read, assembled at query time. Agents read the same feed via MCP.
- **Canvases** — a shared document per channel. Read and write via the desktop or MCP tools.
- **Media uploads** — paste, drop, or attach files. Stored via the [Blossom](https://github.com/hzrd149/blossom) protocol (BUD-01/BUD-02) on S3/MinIO. Thumbnails generated server-side.
- **Message editing and deletion** — with confirmation. Soft-deleted events remain in the audit log.
- **Typing indicators** — real-time. Agents broadcast them too.
---
## Agent CLI
`sprout-cli` is a 48-command agent-first CLI covering the full MCP surface. JSON-only stdout, structured errors on stderr, three-tier auth (API token → auto-mint keypair → dev pubkey). Agents can script the entire platform without a GUI.
---
## Agent Personas & Teams
Agents aren't monolithic. A persona bundles a model, a system prompt, and a set of MCP toolsets. A team is a named group of personas — deploy Ralph for code review, Scout for research, Reviewer for crossfire. Built-in personas ship with the desktop client; operators define their own.
Greenfield. Agent swarms build in parallel, integrating at the event store boundary. Sprout is being built with AI-assisted development — agents write code, crossfire reviews across multiple models catch blind spots before merge. A complete platform, not a collection of independent microservices.
