Agentic automation

Let agents operate the workflow, not take custody of the archive.

VoxNotch gives agents and scripts a concrete surface: upload media, poll jobs, inspect transcripts, run recipes, send outputs, and receive webhooks from the instance you operate.

Agent surface: HTTP API, workflow runs, recipes, jobs
Credentials: Personal API keys with scoped access
Boundary: LLM providers are opt-in egress
Best fit: Private automations and review loops

Agent loop

Give an agent a bounded job with visible outputs.

Ingest and classify

An agent can upload or watch for media, wait for local transcription, tag the result, and place it into the right collection.

A clay automation pipeline connecting API token, local server, and recipe cards. — Automation pipeline with API token, local server, and recipe cards.

Run repeatable recipes

Workflow recipes turn common jobs into repeatable runs: manual clip render, podcast production loops, exports, and webhook notifications.

Automation recipes

Agent-friendly surface

shell

curl -sS "$VOXNOTCH_URL/runs" \
  -H "X-API-Key: $VOXNOTCH_API_KEY"

curl -sS -X POST "$VOXNOTCH_URL/workflow-recipes/podcast_studio_production_loop/run" \
  -H "X-API-Key: $VOXNOTCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"recording_id":"rec_...","execute":true}'

The interface is plain HTTP today, which makes it easy for agents, shell scripts, and internal tools to call without a hosted control plane.

Boundaries

Agentic does not mean unbounded.

API keys are scoped. Webhooks and S3 destinations are explicit egress. Hosted LLMs are opt-in. The useful agent pattern is a bounded worker that leaves an inspectable run behind.

Developer surface

Start with one agent-run recipe.

Have an agent upload one file, wait for transcription, run a recipe, and report the output path.

Join Beta Searchable archive