AI agents, answered
What is Belna Arche 1.0?
Arche 1.0 is Belna's personal AI agent — built on the open-source Kimi K3 model with an agentic harness optimized for privacy and safety. One person claims each agent, names it, and works with it through chat, canvas and vault.
How do Belna AI agents keep me safe?
Three layers: secrets stay sealed in your vault (the agent only gets masked references), sensitive actions pause for your approval, and browser, code and computer use run in a sandbox with a visible trace. See the security page.
What can my agent actually do?
Research with live cited sources, build pages and tools, review GitHub pull requests, triage admin, and remember your preferences across chats. If you can think it, your agent can make it real.
How much does it cost?
Free includes $10 of API costs. Pro is $30 with $20 credit + a $50 gift card; Max is $50 with $50 credit + a $100 gift card. Full details on the pricing page — upgrades are requests until payments connect.
Can I sign in with Google?
Yes. Besides email + password you can continue with Google or use a passwordless email one-time code. All methods run through Supabase Auth and land on the same account with the same credit and vault.
What is Arche 1.0 built on?
Kimi K3 open weights (2.8T-parameter MoE, 1M-token context, native vision) served through the Belna harness. Real benchmark numbers are on the models page.
Where is my data kept?
Your account and vault live in Supabase with encryption at rest; secrets are masked everywhere and the model only receives references. Under GDPR you can access, correct or delete your data anytime — see privacy.