Skip to content

Blackrim Vox

Voice capture, routed to where it belongs

Capture spoken intent from any source — your own voice, an in-person meeting, an online call — and route it to the right sink: an LLM with the right context, S3-compatible storage, your inbox, or all three. Vox treats the microphone as a first-class input to your work, not a transcription afterthought.

Pre-alpha

Blackrim Vox is under active development. APIs, config schemas, and CLI flags may change between releases without a deprecation window until v1.0. See the status & roadmap below.

Why Vox

  • Self-hosted, BYOK, no proxy. The Go binary runs on your machine. Your provider keys never leave your environment, and Vox does not call home.
  • One pipeline, three sources. Self / in-person / online calls share the same capture → segment → ASR → route → sink stages — source-specific code lives only at the capture edge.
  • Structured intent, not a transcript dump. The router classifies each segment (question, command, follow-up, todo) and frames it for the right destination — a model prompt, a slash command, a bd issue, a meeting summary.

What it does:

  • Routes speech through pluggable ASR, LLM, and TTS backends under a single operator-controlled pipeline.
  • Enforces role-based access gates and emits structured audit events to Splunk, Datadog, syslog, Loki, or Elasticsearch.
  • Ships as a single Go binary with no mandatory cloud dependencies and no default telemetry.

What it does not do:

  • Manage your LLM keys or proxy your traffic — you bring your own keys (BYOK) and the binary stays on your infrastructure.
  • Provide a managed service, hosted dashboard, or SLA — those belong to the enterprise tier (see Editions).
  • Make architectural decisions for you — pipeline stages are config-driven and replaceable.

Try it in 60 seconds

# 1. Clone and build
git clone https://github.com/Blackrim-Vox/blackrim-vox.git
cd blackrim-vox
make build            # produces bin/vox

# 2. Start a local session with the echo ASR backend (no keys needed)
./bin/vox listen --asr=echo --sink=local-file

# 3. In another terminal, send an audio frame
echo "hello vox" | ./bin/vox transcribe --asr=echo -
# Prints the echo transcript to stdout and appends to the local-file sink.

No API keys, no network calls, no data leaves the machine.

Explore the docs

Positioning

Blackrim Vox is open-core, Apache-2.0 licensed. The core binary — ASR routing, LLM gating, TTS dispatch, RBAC, and all forwarders — is free and open source. There are no usage limits, no telemetry beacons, and no call-home in the open edition.

BYOK (bring your own keys). The binary reads API credentials from environment variables or a local config file. Keys never leave the operator's environment.

Per-operator scope. Each deployed instance is a single logical operator. Multi-tenancy, SSO federation, and centralized key management are enterprise-tier features.

For the enterprise tier (pooled credentials, WorkOS SSO, SLA support), see Editions.

Status

Blackrim Vox is pre-alpha with a working end-to-end pipeline. The capture → segment → ASR → router → sink path is production-ready for single-operator dictation and transcription today; multi-tenant, hotkey, and mobile arcs are in flight.

Until v1.0, APIs, config schemas, and CLI flags may change between releases without a deprecation window.