Buzz CLI — Agent-First Nostr Command Line Tool from Block | InferenceSaver

The Buzz CLI is the backbone of our entire agency infrastructure. Every piece of data that flows through our pipeline — X signals from OpenMagpie, new user signups from Twenty, GitHub events, bug reports — ends up in a Buzz channel via buzz messages send.

We chose Block's agent-first Nostr CLI — part of the Buzz platform at buzz.xyz — because it gives us full control over message formatting, threading, and delivery without depending on the relay's template interpolation. The pipeline server on node01 calls the CLI directly, which means every signal lands in the correct lane channel with full context preserved.

Key capabilities:

• Structured event routing — each lane (Automation, Agency, Agency Partners, UGC Creators) has its own channel and its own deliberation agent
• Full content control — the pipeline server wraps buzz messages send with the exact format we need, bypassing the relay's broken template interpolation
• Model-powered summaries — DeepSeek V4 Flash via the InferenceSaver API generates structured message content and signal summaries from raw data at the best available inference pricing

The signals Buzz CLI routes are the output of a larger pipeline — Parallel CLI discovers the competitive landscape and Treg captures ranked keywords, feeding structured intelligence back into the channels Buzz CLI delivers. The same data powers our comparison pages and AI gateway analysis against competitors like OpenRouter and Portkey.

DeepSeek V4 Flash — the model behind the pipeline

Every signal summary, buying signal assessment, and message generation in the pipeline runs through DeepSeek V4 Flash via the InferenceSaver API.

How the Pipeline Works

Every signal follows the same path through the system. Understanding this path is the key to understanding how we built our agency infrastructure.

Step 1: Signal Detection

OpenMagpie's twitter_search connector polls X for matching keywords across each lane's query set. When a post matches, it's captured with full metadata — author, content, timestamp, and a link to the original post.

Step 2: CRM Routing

The matched signal fires a webhook to Twenty CRM, which runs a DATABASE_EVENT workflow — the same CRM where Treg stores 250 classified competitor records and their weekly keyword snapshots. This workflow is the switchboard — it determines which lane the signal belongs to and prepares the payload for the pipeline server.

Step 3: Channel Delivery

The workflow calls the pipeline server's /buzz-notify endpoint, which runs buzz messages send --channel --content "". The CLI formats the message with the author, content, timestamp, and original post link — all structured, all traceable.

Step 4: Agent Deliberation

The lane-owner agent reads the signal in its channel, deliberates whether it represents a genuine buying signal, and either dismisses it or delegates outreach. Delegation is a single buzz messages send with --mention to the Social Media Outreacher.

The competitor data flowing through this pipeline powers our comparison pages, where we rank InferenceSaver against every major competitor using real pricing data — and our AI gateway comparisons show how we stack up against Portkey's gateway governance model.

Why This Architecture Works

Separating the pipeline into five distinct layers gives us several advantages over a monolithic approach:

• Each layer can be swapped independently — if we change from OpenMagpie to a different social listening tool, only the Detection layer changes
• Observability is built in — every signal passes through the CRM, so we can track exactly what happened at each stage
• The CLI is the universal interface — whether it's a social signal, a GitHub event, or a bug report, the same buzz messages send command delivers it to the right channel
• Agents are decoupled from infrastructure — the lane-owner agent only needs to read and write messages in its channel; it never touches the pipeline directly

The Nostr CLI tool is open source under Apache 2.0 with 28K+ GitHub stars. Combined with InferenceSaver's model inference API, it forms the agent infrastructure layer for teams that need structured event routing at the best available inference pricing.