Now in Open Beta — Free Self Hosted

Talk to your backlog.
Watch it run itself.

Voxyflow turns conversations into cards, and cards into shipped work — while you sleep.

See it in action.

From a single prompt to shipped work — no edits, no script. One real session.

Features you wish it existed

Three core capabilities working together so your project manages itself.

Conversational Kanban

Stop clicking through UI. Just talk. Say "move the auth task to in-progress" or "show me everything blocked" — Voxyflow understands your intent and updates your board instantly. Your backlog becomes a conversation.

  • Create, move, and close cards by chat
  • Natural language queries across your board
  • Works in any language, any style
Core

Orchestrated Execution

Voxyflow doesn't just track work — it does it. Spawn AI worker agents on any card with one command. They research, write, code, and deliver right inside your project. You review; they execute.

  • 7 specialized agent types
  • Full task lifecycle — plan, execute, report
  • Smart model routing — Opus, Sonnet, or Haiku, auto-picked per task

Autonomy

Set a directive, walk away. The Autonomy system wakes up on a schedule, reads your project state, decides what to tackle next, and ships it. Wake up to progress reports, not status meetings.

  • Cron-based autonomous cycles
  • Reads board state before every run
  • Pause, steer, or stop anytime

And so much more

  • Custom personality — not a faceless chatbot
  • Per-project RAG + memory system
  • Knowledge Graph + ChromaDB + session memory
  • Native voice — STT + TTS, openWakeWord support
  • PWA — install on phone, mobile-friendly
  • Light token usage — smart routing keeps costs low
  • MCP Tools
  • API Endpoint

The possibilities are infinite.

From zero to autonomous in 3 steps

Self-host the beta. Bring your own AI key. Then just talk.

1

Clone & run

Voxyflow is open source (AGPL-3.0) and runs entirely on your machine.

git clone https://github.com/jcviau81/voxyflow

Follow the README — it's a beta, install is hands-on, but you'll be up in a few minutes.

100% local · Open source · No telemetry
2

Connect your AI provider

BYOK, provider-agnostic. Paste a key for OpenAI, Anthropic, Gemini, or point to a local Ollama — Voxyflow talks to it directly.

Already on Claude CLI? Even simpler:

claude login

Voxyflow picks up the session — no key to paste.

Whatever you connect, credentials stay on your machine. No proxy, no telemetry, no middleman.

OpenAI · Anthropic · Gemini · Ollama
3

Chat your board, then walk away

Talk and brainstorm your project. Create cards, refine your tasks and features. When you're ready to step away, write a directive — "Keep shipping from Todo, prioritize bugs" — flip on Autonomy, and let the magic work.

You: "Create a card: write blog post about launch, assign to marketing, due Friday."
Voxyflow: ✅ Done. Card added to Todo.

Voxyflow wakes up every few minutes, picks the next task, and executes.

Autonomy active · Last run: 2 minutes ago

Free now. Hosted later.

Self-host today, free forever. A managed Hosted version is on the roadmap — pending the right LLM partnership.

AVAILABLE NOW
Beta · Self-hosted
$0
Forever · BYOK · AGPL-3.0
  • All features unlocked
  • Runs entirely on your machine
  • BYOK — you bring the AI key
  • Open source, AGPL-3.0
  • Community support
  • No telemetry, no lock-in
Get it now
Pro · Hosted
TBA
Managed instance · LLM bundled · No install
  • Everything from self-hosted, plus:
  • We host it for you — no server, no install
  • LLM access bundled (no separate subscription)
  • Backups & updates handled
  • Pending the right LLM partnership

Want to discuss enterprise self-host or partnership? Email me.

Need support or want to contribute?

Reach me on GitHub or Discord.

Frequently asked questions

Everything you need to know before joining the beta.

Voxyflow ships calibrated for Claude — Sonnet for implement, Haiku for quick tasks, Opus for architecture. Run claude login once and you're set, no key paste needed.

Need a different setup? Worker Classes let you remap any task type to any model. The system is provider-agnostic by design, with Claude (via CLI) and local Ollama being the most thoroughly tested combos today. OpenAI, Gemini, and other providers work — your mileage may vary on quality per task type.

BYOK throughout — you bring the keys, Voxyflow routes the work.

Yes — that's the only way to run Voxyflow today. Clone the repo, run it locally, you're up. There is no SaaS or hosted version. A managed Hosted variant might come later (pending the right LLM partnership), but the self-hosted path will always remain free and open source.
Everything stays on your machine. Your project data, conversations, AI credentials, and worker outputs live in your local install — nothing is sent to a Voxyflow server, because there is no Voxyflow server. When you make AI calls, they go directly from your machine to your provider of choice. No proxy, no telemetry, no middleman.
The self-hosted version stays free, open source (AGPL-3.0), and BYOK — forever. "Beta" just means the project is young and moving fast. If we ever launch a managed Hosted version, it'll be a separate paid offering — your local install remains untouched and free.
No. AGPL applies to Voxyflow itself — meaning if you redistribute Voxyflow or host it as a network service for others, those modifications need to be open. Your projects, code, and outputs created with Voxyflow remain entirely yours, under whatever license you choose. AGPL is about the tool, not what you build with it.

Autonomy lets Voxyflow keep working while you don't. You write a directive — something like "Keep shipping cards from Todo, prioritize bugs" — flip the Autonomy toggle on, and Voxyflow wakes up on a schedule, reviews your project state, picks the next task, and executes it with a worker agent. Come back later to approve the results.

You can adjust the schedule, pause/resume the toggle anytime, and watch live worker output as it happens — no black-box agent.