Codec is the open-source runtime Lucy, Intake, and Beacon run on. 245 features · 57 skills · 378 tests · 34K+ lines of code. You can self-host, fork, ship your own agents on top — or just read how we built it. We made it open because renting agent infrastructure shouldn't be a business.
A short tour of what Codec actually does — the framework behind every AVA agent, running in the open.
Codec is the runtime. Around it we ship six surfaces — voice, dictation, instant agents, chat, vibe sessions, voice-first — plus an overview surface that ties them together. All open source. All swappable. All shipped from the same monorepo.
The runtime. Skills, planner, evals, tool-use. Everything else extends from this.
Push-to-talk dictation. Voice in, structured text out, into any field on any app.
Spin a one-shot agent on demand. No setup, no project file. Ask, get, dismiss.
Persistent conversational surface. Memory, context, your skills, your tools.
Sessions tuned for creative flow. Pair with a model that matches your taste.
Hands-free, voice-in / voice-out. Drive your machine without touching the keyboard.
Single pane of glass across every surface. See, switch, and orchestrate every Codec session you have running.
One drag-and-drop install. The full Codec suite, packaged for macOS. Bring your own AI brain — we never resell tokens.
Built for Apple Silicon. Drag-and-drop install. Menu-bar, hotkeys, system-wide voice. No browser, no Electron tax.
Plug in Anthropic, OpenAI, Gemini, or a local model. Switch per surface. We never resell tokens — you pay the model provider directly, plus 20% routed through us.
Push-to-talk anywhere. Dictation, conversation, voice agents. The keyboard becomes optional.
Self-host the runtime. Build your own surfaces. We don’t see your traffic.
+ AI brain at provider cost × 1.20 (we route, never resell). Cancel anytime, OSS still yours.
We spent two years watching the agent tooling space congeal into closed platforms — each of them trying to be the landlord of your workflows. We don't think that works long-term. The interesting surface area is above the runtime, not inside it.
Codec is our bet that the best agents will be built on open foundations — and the best agencies will be the ones who can run yours end-to-end. You pay us for the people and the playbook. Not the plumbing.
Self-host on your cloud. We never see customer data unless you ask us to.
Agents and skills are portable. If we disappear tomorrow, you keep running.
Build your own agent on Codec. Sell it, license it, hide it. No strings.
Every skill, every prompt, every eval. If you can't audit the agent, you shouldn't run it.
Every Codec install is a migration out of the SaaS subscriptions that stack up every month. Same tools. Same job. No monthly bill after setup.
Download from the App Store. Allow permission. Done. No terminal. No dev account. No API keys. We handle all the infrastructure — you just run the app. Usage billed monthly, operational cost included.
The simplest way to bring real AI to your Mac.
We source a Mac Studio, pre-configure every open-source model you need (Qwen, Llama, Mistral, Whisper, Stable Diffusion — your choice), customize it to your workflow, and ship it ready to plug in.
Real AI ownership. Your hardware, your models, your privacy. Zero cloud dependency. Zero subscription after install.
A working voice receptionist — hooked to your calendar, branded, deployed. Walk through the codec tutorial in under 10 minutes.
Drop a star, open an issue, ship a skill. The framework is free forever — the agency around it is how we eat.