One engineer,
the output of an entire team.
Vodou OS is your agent harness for AI execution.
Your plate is full — more projects, more tools, more context than one person can hold. Vodou doesn’t help you work harder. It lets one person carry more work at the same time: every request routed to the right model, skill or tool, run in parallel, with everything you’ve already done still in memory.
Meet Vodou.
Let’s accelerate what the world’s boldest AI builders can create.
Try Vodou OS free for 60 days.
Download Vodou today and use the full product, free, while in Alpha. Contribute to the repo for lifetime access! After that, pick a subscription to keep going. No credit card required up front. See pricing below.
- Full product access while in alpha
- Alpha-tier pricing locked in if you subscribe
- Your feedback shapes the roadmap
By downloading you agree to the Alpha Terms. When you sign in, Vodou reports usage metadata (token counts, model and tool names, timings and errors) to your Vodou account for your usage dashboard and billing. Bring-your-own-key users can turn this off in Settings.
The work keeps growing. The day doesn’t.
One engineer is now expected to ship what used to take a team — build, review, document, support, report. AI was supposed to absorb that load. Instead it added a second job: deciding which model, prompting it, re-explaining context, stitching the output together. The ceiling isn’t your skill. It’s how much you can hold at once.
Capacity is the bottleneck
Work arrives in parallel and you execute it serially. Throughput — not effort — is what caps your output and what it’s worth.
Context resets constantly
Every new tab means re-explaining the project. Memory that doesn’t compound is memory you pay for twice.
Fragmented, and off your machine
Models, tools and data live in separate silos — and most of it leaves your hardware to get anything done.
Plain English in. Executed answer out.
The BrainLoader — Vodou’s proprietary router — reads intent, loads memory, and picks the right skill, tool or model before a token of inference is spent.
One harness that controls every model and every tool.
Vodou OS installs on your machine and becomes the conductor for everything AI touches in your day. Memory, routing, execution, observability — all in one place.
A brain that doesn’t forget.
Persistent, structured memory of your files, projects, and conversations — searchable, private, and yours.
Run every model. Pick the winner.
Parallel calls to local and frontier models, with automatic routing, fallback, and quality scoring. Stop guessing which model to use.
See Vodou in motion.
A first look at the dashboard, the orchestration views, and the live execution surface.
Everything you need to operationalize AI.
Running on a working Alpha install today: unlimited skills, unlimited MCP servers, unlimited intent mappings, unlimited memory and unlimited delivery channels — on your own machine.
Deterministic routing
The BrainLoader resolves unlimited intent mappings to skills, tools and scripts before any inference happens — not after the model guesses.
Local-first execution
Runtime and memory are 100% local, with llama.cpp bundled and LM Studio and Ollama supported, so inference can stay on your hardware too.
Compounding memory
Unlimited memory chunks with full-text + vector retrieval, a reranker, and contradiction and dedup passes. Governed, not a chat log.
Memory that follows you
The Bridge extension captures 22 AI chat surfaces and injects your memory back into 18 verified ones — plus Cursor and Claude Code.
Parallel tool execution
Connection pooling dispatches skills, MCP tools and scripts concurrently — 25–50× faster than running them one at a time.
Autonomous kanban
Drop work on the board and Vodou’s agents pick it up and run it — scheduled tasks, triggers, hooks and cron always on.
Composable skills & MCP
Unlimited skills and MCP servers. Vodou is also an MCP host — any local client attaches with its own token and vault.
Use it from anywhere
Slack, Teams, Telegram, Discord, WhatsApp, Signal, iMessage, Google Chat, voice and web — unlimited channels, you control who’s allowed in.
Governance by default
Zero mandatory cloud. Per-client identity, pre-dispatch egress profiles, salted audit log, rate limits, and surgical revoke.
Connected to the apps you already live in.
Vodou pulls context from your tools and brings results back where you already work. No exports, no copy-pasting, no context switching.
// Logos shown are trademarks of their respective owners.
Connectors for real work.
One Agent.
Every AI Model.
Vodou is not locked to one provider. It can call Claude, GPT, Gemini, Kimi, Mistral, and DeepSeek as the job changes: reasoning, coding, long context, research, open weights, or cost.
// Skills tiles will join this grid as the skill library grows.
Everyone owns one layer. We own the harness between them.
The field has split into three camps: agent runtimes that execute, workspaces that aggregate models, and memory add-ons that remember. Each is strong in its lane and blind to the other two. Vodou is the only one that runs all three on your own machine, under one governed pipeline.
| HORIZONTAL LAYER | AGENT RUNTIMES | MULTI-MODEL WORKSPACES | MEMORY ADD-ONS | |||||
|---|---|---|---|---|---|---|---|---|
| CAPABILITY | VODOU public alpha | CLAUDE COWORK Anthropic · Opus 5 default | OPENCLAW 2026.6.8 stable · ~379K★ | HERMES AGENT v0.20.0 · Aug 3 2026 | POE Quora · $20.99/mo Pro | ANYTHINGLLM MIT · desktop + docker | MEMORYPLUGIN $79–149/yr · 21+ tools | BYTEROVER free / $14.90 · IDE-side |
Deterministic routing Before the LLM | ✓BrainLoader — unlimited intent mappings resolve skills, tools and scripts before inference | ✗Claude-native agent loop decides | ~Skill match, then the model decides | ~Agent loop over self-distilled skills | ✗You pick the bot | ~Rule-based model routing only | ✗Not a runtime | ✗Not a runtime |
Local-first execution Incl. inference | ✓Runtime and memory 100% local; bundled llama.cpp for local inference, plus LM Studio and Ollama | ~Agent loop and file access local; shell and code run in Anthropic’s VM, inference cloud | ~Self-hosted gateway, cloud LLMs | ~Self-hosted; serverless execution path | ✗Fully cloud | ✓Fully local with Ollama backend | ~Hosted by default; local MCP server offered | ✓Local by default, cloud sync optional |
Compounding memory Governed, not a log | ✓Unlimited memory — FTS5 + vector + reranker, contradiction and dedup passes, health-gated retrieval | ~Memory and consolidation — cloud-held, not inspectable | ~Plaintext markdown diaries, unencrypted at rest | ~Full-text recall + user modeling (off by default) | ~Off by default, updates once a day, absent in group chat | ~Per-workspace RAG; no cross-session user model | ✓Buckets, edits, history import — vendor-held | ✓Hierarchical store, high retrieval accuracy claimed |
Memory that follows you Across other AIs | ✓Bridge extension captures 22 chat surfaces and injects back into 18 verified — plus Cursor and Claude Code | ✗Anthropic surfaces only | ✗Its own agent only | ~Pluggable memory providers — still only inside Hermes | ✗Poe only | ✗Its own workspaces only | ✓21+ AI tools — the closest direct rival | ~IDE agents only (Cursor, Copilot, Claude Code, Cline) |
MCP Client and host | ✓Unlimited servers as client, and a host: any local MCP client attaches with its own token, profile and vault | ✓Client — local and remote connectors | ✓Client | ✓Client | ~No first-party MCP; community servers wrap the API | ✓Client, and serves as a server to Claude Desktop | ✓Hosted remote MCP (OAuth/PKCE) + local server | ✓MCP into any AI IDE |
Parallel tool execution | ✓Connection pooling, 25–50× over serial dispatch | ~Parallel sub-agents on Opus 5 | ~Serial by default, sub-agents available | ~Agent loop plus agent-to-agent | ~Side-by-side model compare, not tool orchestration | ~Sequential agent runs | ✗— | ✗— |
Composable skills | ✓Unlimited skills, user-authored, no registry gatekeeper | ~Skills and plugins, curated by Anthropic | ✓Community skill hub — 13.7K skills | ✓Self-distilled skills + plugin SDK | ~Custom prompt bots, no tool runtime | ~Agent builder over MCP tools | ✗— | ✗— |
Skills that improve themselves learning loop | ✓Drafts new skills from how you work; you review and promote them, so nothing changes without your approval | ✗Static skill library | ✗Skills are hand-installed | ✓Built-in learning loop — the agent authors and refines its own skills | ✗No skill runtime | ✗Static agent config | ✗— | ✗— |
Team & org controls collaboration, admin, spend | ~Per-client identity and revoke today; org-wide admin ships with Vodou Enterprise | ✓Org feature and spend management, shared review flows | ~Self-hosted, admin is yours to build | ~Multi-agent delegation, no org admin layer | ✓Group chats up to 200 people across 200+ models | ~Workspaces with per-user permissions | ✗Single user | ✗Single developer |
Multi-channel delivery | ✓Unlimited channels — Slack, Teams, Telegram, Discord, WhatsApp, Signal, iMessage, Google Chat, web, voice | ✗Desktop, web and mobile app only | ✓25+ platforms through the gateway | ✓20+ channels, streaming voice in v0.20 | ✗Poe app only | ✗App only | ✗Browser extension only | ✗IDE only |
Model freedom BYOK and local | ✓OpenAI-compatible API, any provider, local models first-class — BYOK never metered | ✗Anthropic models only | ~Any provider you configure | ~Multiple providers supported | ✓Every major model on one subscription — but only theirs, chosen per chat | ✓Any provider, with dynamic routing rules | ✗— | ✗— |
Governance Audit, scope, revoke | ✓Zero mandatory cloud. Per-client identity, egress profiles enforced pre-dispatch, salted audit log, rate limits, surgical revoke | ✗Cowork activity excluded from audit logs, compliance API and data exports — every tier | ✗60+ CVEs, ~30K exposed instances, a poisoned skill marketplace | ~Self-hostable; secrets via 1Password / Bitwarden | ✗Vendor-held, no tenant controls | ~Self-hosted, but no governance layer of its own | ~Encrypted in transit and at rest — vendor-held | ~Local by default, cloud sync opt-in |
Unified pipeline One product, not four | ✓Routing + memory + tools + delivery in a single governed path | ✗Agent only | ✗Agent + skills, no unified pipeline | ~Agent + learning loop, memory pluggable not governed | ✗Model access only | ✗RAG workspace only | ✗Memory only | ✗Memory only |
The people already running it daily.
Vodou has been running with a small group of design partners — operators, builders and technical leads using it against real workloads, not demos. Their feedback is what set the Alpha feature lock.
The biggest benefit of Vodou is capacity. I can manage more complex work at once without spending so much time coordinating between different tools.
Vodou helps me pick up where I left off without constantly re-explaining the context. I’m able to stay in motion and move my work forward much faster.
Vodou lets me focus on the outcome instead of figuring out which model, tool, or workflow to use. It removes a lot of the manual coordination from AI-assisted work.
Instead of moving through everything one task at a time, Vodou helps me advance multiple projects in parallel. I’m getting more leverage from the same amount of time.
Connection pooling runs skills, MCP tools and scripts concurrently instead of one after another.
Full-text + vector retrieval with contradiction and dedup passes, gated on memory health.
User-authored skills and connected tools, orchestrated automatically — no registry gatekeeper.
// Figures are alpha-build usage numbers, not customer outcome claims.
Two ways to run. One simple price.
Bring your own model keys and pay your provider directly, or let Vodou cover the model costs for you in one monthly price. Every plan ships with a full 60-day free trial during Alpha — no credit card required. Download the Alpha.
BYOK
The complete OS on your own model keys. You pay us for the platform; your model provider bills you directly for usage.
Bundled LLM & tokens
Model usage is included — the fastest way to start without setting up provider accounts or managing API keys.
Running Vodou across a team or a whole department? Implementation engagements are handled through Vodou Enterprise.
Operationalize AI across the whole org.
Most organizations adopt AI in fragmented, uncoordinated ways — departments operate independently, workflows stay disconnected, institutional knowledge gets trapped. Vodou Enterprise creates one coordinated execution layer across the business, built on the same runtime powering Vodou OS.
Talk to us about Enterprise- Operationalize AIEmbed it into the workflows your business actually runs on, not a side-channel chatbot.
- Preserve institutional memoryKnowledge that survives turnover instead of staying buried in people’s heads.
- Centralized governanceGuardrails, permissions, audit and private or fully local deployment options.
- Custom execution systemsBuilt around how your organization actually operates, on the Vodou OS backbone.
- Phased engagementDiscovery, design, pilot, deployment, then scale across the org.
Questions, answered.
What’s included in the Alpha?+
The full Vodou OS app for macOS, Windows and Linux — orchestration, memory, tool graph, and multi-model routing. Some advanced enterprise features (SSO, audit logs, team workspaces) will land before general availability (GA).
What happens after the 60-day trial?+
You’ll be prompted to choose a subscription to keep using Vodou OS. Alpha users get founding-member pricing locked in for the life of their subscription.
Which platforms does Vodou run on?+
macOS, Windows and Linux, all available now in Alpha: macOS 13+ on Apple Silicon or Intel, Windows 10/11 on x64, and Linux on x64 or ARM64. Each installs with one command from the download section above.
Is my data sent to the cloud?+
Your memory, files, and tool graph are stored on your machine, not in our cloud. Model calls go directly from your machine to whichever provider you configure, or to your local models. When you are signed in, Vodou also reports usage metadata (token counts, model and tool names, timings and errors) to your Vodou account for your dashboard and billing; bring-your-own-key users can turn that off in Settings.
Which models does it support?+
All of the major ones, and you can switch per task. Included with Vodou-powered plans: Vodou LLM, with no API keys to manage. Bring your own key: Anthropic (Claude), OpenAI, Google Gemini, xAI (Grok), Mistral, DeepSeek, Kimi (Moonshot), Groq, Fireworks, Together and OpenRouter, which opens up hundreds more models. Use a subscription you already have through the Claude CLI or Kimi Code CLI. Run fully local with Ollama, LM Studio or Vodou Local (bundled llama.cpp). And any OpenAI-compatible endpoint.
Is this open source?+
Vodou is open core. The client and orchestration stack (the gateway, MCP servers, browser extension, skill format and installers) is open source under Apache-2.0 at VodouAI/OS. The Rust engine that runs routing and memory is proprietary and ships as a signed binary from VodouAI/vodou-core under the EULA. The ecosystem around it is open for contributions: skills in VodouAI/vodou-skills-catalog, app presets in VodouAI/Apps, and community lenses in VodouAI/lenses-directory.
Let’s accelerate what the world’s boldest AI builders can create.
Download the Alpha, run it against your own work for 60 days, and tell us what to build next.