Clawgo

Why it's ranked here

Clawgo has a focused job and exposes the useful controls directly. Pairing, session routing, chat subscription, speech input adapters, local text-to-speech, service discovery, and provider delivery are all present. Its small surface suits dedicated nodes, though it depends on a separate gateway and external speech tools.

What's good

Transcript routing is explicit and extensible through registered policies. Empty input is ignored, bounded queues reject excess work instead of blocking, and tests cover full queues, closed queues, direct transcripts, agent requests, and Telegram quick actions. Cross-compilation instructions make Raspberry Pi deployment practical.

Tradeoffs

This is infrastructure for an existing gateway, not a standalone assistant. The documented deployment requires building a Go binary and may require external programs for speech recognition or synthesis. Its built-in quick action only recognizes a Telegram ping, and the default system voice depends on espeak-ng.

How to use it well

Use Clawgo for an always-on Linux or Raspberry Pi voice terminal connected to an established gateway. Feed finalized transcripts through a FIFO or the supported speech recognizer, then subscribe to replies for spoken output. Choose it for a service-managed headless node, not for a graphical client, gateway replacement, or complete voice assistant stack.

Observed

Primary language

Go

Install surface

Build from source with the Go toolchain into a single CLI binary.

Interface

Command-line application with pair and run commands.

Platform support

Documented for Raspberry Pi and Linux, including Linux ARM64 cross-compilation.

Discovery

Advertises node presence through mDNS using zeroconf.

Test structure

Repository text includes unit tests for routing policy and bounded queue behavior.