Three interlocking layers: a local-first AI agent that indexes your web context, hardware-bound cryptographic identity, and a bring-your-own-model inference layer that never touches our servers.
The Dart Agent runs entirely on your device. It indexes your tabs, bookmarks, history, and local PDFs into a semantic store. Ask it anything; it retrieves from what you’ve actually read, not from the open internet.
Ask the agent to compare, contrast, or summarise across your open tabs and recent history. It cites sources, shows its work, and runs without sending a character to our servers.
Chain actions: navigate to a page, extract a value, fill a form, submit. The agent executes agentic workflows within bounds you set per domain and per profile.
The semantic index persists across sessions. Come back to a topic weeks later and the agent recalls the context from your past browsing — not a cloud service’s index of you.
Hardware-attested identity means your private key is generated inside your device’s secure hardware module — a TPM 2.0 on Windows/Linux or Secure Enclave on macOS — and never exported. Not by you, not by us, not by anyone.
Private keys live inside the hardware security module. Signing operations happen in the chip. Keys cannot be exported even with admin rights.
NIST-standardised post-quantum algorithms for key exchange and digital signatures. On by default for all new sessions. Hybrid mode for compatibility with classic TLS endpoints.
Dart is the only browser that makes passkeys and FIDO2 hardware authentication a first-class UX rather than a buried setting. Passwords are the fallback, not the default.
Dart connects to any OpenAI-compatible API endpoint. Your API key never reaches our servers. Your prompt traffic goes directly from your browser to your chosen provider, with no relay, no markup, and no logging by us.
API keys are stored in macOS Keychain, Windows DPAPI, or libsecret on Linux. They are never written to our database or network traffic.
Your inference requests go directly from Dart to your provider. We are not in the path. There is no relay, no logging, no cost to us, no cost to you beyond your provider’s rates.
Auto-detects Ollama and LM Studio on localhost. Runs fully offline. Zero bytes to any external server. The only architecture that satisfies regulated-industry requirements.
Point Dart at your organisation’s LiteLLM, Bedrock, or Vertex gateway. Team members sign in via SSO; the gateway enforces your org’s model policies.
Private beta is running now. Join the waitlist and we’ll send you a personal invite.
No commitment · we’ll write back personally