What it does
What PAVONA does
Chat with local or cloud models, give each agent its own tools and permissions, and keep a monthly cap on paid calls. Take the tour for five workflows, step by step, or read about conversations, ongoing work and companions.
Choose your model.
Chat locally or through a cloud provider. History stays on your disk; connected services receive the content they need.
Model guide ↗Shape each agent.
Set its model, tools, instructions and permissions. Review those settings before it acts on your files.
Permission guide ↗Cap cloud spending.
Paid calls need a monthly budget. PAVONA warns at 80% and stops at the cap. Billing can lag; check your provider account too.
Find more models.
Browse Ollama and OpenRouter. Availability depends on your provider and account. Local inference has no provider fee.
You choose when to update.
The open app checks every six hours. You approve installation.
Patch notes ↗Recorded benchmark
Read results with their limits.
PAVONA tasks passed: 94% · 2026-09-24.
Exploratory run. Matching model and resource budgets are unverified; this does not establish superiority or predict your results.
Methods and all results ↗Privacy
Know where your data goes.
Local models run here. Cloud models and connected tools send data to your chosen services.
Read the network details
Research, model downloads and update checks also use the network. Local storage is accessible to other software on your computer. Data details · The privacy page.
Check it yourself
Inspect the source.
Source and tests come with the download.
python run-tests.pyTests may use models, network or files. This page does not claim a passing run.
Open the handbookNow playing in the corner: Feast Your Eyes by Mastodon