Vocion docs
The open framework for production-ready agents and agentic workflows. Built-in retrieval, typed plugins, MCP-native, full observability. Self-hostable. No vendor lock-in.
What problem we solve
The hard parts of taking an agent from prototype to production aren't the model calls — they're everything around them: connecting to your data, chunking it, embedding it, retrieving it, caching it, observing it, governing it, evaluating it. Vocion ships those primitives as a typed, pluggable framework. You bring the data and the prompts; we own the plumbing.
Get started
Pick the path that matches what you want to do.
| You want to… | Read |
|---|---|
| Run Vocion locally + ship your first Skill in 10 min | Quickstart |
| Walk a real end-to-end workflow | Worked example: discovery follow-up |
| Deploy to production | Deployment |
| Connect a new data source | Sources |
| See everything your team did | Dashboard → Activity (one feed: checks, runs, fires, syncs) |
| Build a custom plugin | Writing a plugin |
Foundations
The five things you author. Read all five (~15 min) and you understand Vocion.
- Sources — connected systems that feed raw data in. Ingestion, embedding, retrieval, caching.
- Objects — first-class business entities your tenant cares about.
- Skills — single LLM-powered units of work with typed I/O.
- Workflows — sequences of skills with human approve gates and durable state.
- Missions — open-ended, goal-driven team work; the third work mode.
- Agents — LLM orchestrators that wire skills + workflows + sources into a named identity.
- Teams — agents grouped as a lead + specialists; the org chart of a deployment.
- Automations — the WHEN of the system: schedules + events → run a workflow or check a mission.
- Members — the humans in the account: roles, management, link-based invites (no mail server needed).
How-to
Walk-throughs for recurring tasks.
- Authoring context — edit YAML + markdown, sync to DB.
- Using built-in tools — enable web search, browse, image, code, and artifacts.
- Running a mission — brief open-ended work, watch the team run it, review + promote.
- Evals — fixtures alongside each resource; CI catches regressions.
- Feedback + logs — rate runs, browse the timeline, close the loop.
- Self-hosting — install topology, requirements, local dev.
- Connecting sources — credentials: the Sources UI, the CLI, and durable Google OAuth.
- Extracting a tenant repo — split
workspace/<org>/into its own git repo.
Operate
- Deployment — AWS single-EC2 reference + scaling pressure points.
- Observability — Langfuse stack, trace dimensions, cost slicing.
- Troubleshooting — common failure modes and how to fix them.
Reference
Stable interfaces you'll return to.
- CLI — every
npm runscript. - Glossary — controlled vocabulary for every term.
- MCP tool reference — every tool the MCP server exposes.
- Retrieval config — every knob in
retrieval.yaml. - Repo architecture — the layered model (core / SDK / plugins / starter / install).
- Authentication — auth.js + oRPC auth guards.
API
Programmatic access for external systems — push objects, trigger runs, poll results, register webhooks.
- Overview — API vs MCP vs A2A.
- Authentication — tenant-scoped Bearer tokens.
- Resources — CRUD on Agents, Skills, Workflows, Objects, Sources.
- Runs — trigger, poll, approve, cancel.
- Webhooks — subscribe to completion + approval events.
License
MIT.