Our founder on the benchmark that defined our wedge:
Repeated company workflows compiled into local executable skills; verifiable, auditable, customer-owned execution.
Reliable agents need more than bigger models. They need substrate.
Running a hard live-web workflow on our substrate, a cold prompted browser agent succeeded 8 out of 10 times at a median ~390s.
After we compiled it into an executable skill, the same task hit 28/30 on a fresh local iGPU at ~58s median.
The bottleneck wasn't the model. 🧵
For DeFi auditors:
the source is now public.
SECURITY.md defines the disclosure workflow.
For protocol builders:
MIT-licensed interfaces and libraries are designed for composability and vendor-friendly integration.
For the Base ecosystem:
this is what production-grade digital-asset protection infrastructure looks like.
@base@CoinbaseDev@baselatam
Sovren Software now maintains two flagship public OSS repositories.
Visage
Linux face authentication through PAM.
Rust. v0.3.0.
Mr Haven V1
Programmable time vault infrastructure on Base.
Solidity. v1.0.2.
Different domains.
Same engineering discipline.
System architecture:
• Per-user vaults deployed as EIP-1167 minimal-proxy clones
• Shared yield routing through Aave V3
• User-defined distribution plans
• Scheduled and inactivity-triggered execution flows
• Beneficiaries specified by wallet or email via hash-locked escrow
852+ tests passing.
Vendored dependencies are pinned to deploy-time trees to support deterministic recompilation and bytecode reproducibility.
Voice surface: @openhome | Action surface: Esver OS
Both local. Reliable computer-use agents need substrate and executable skills, not bigger models.
$0 marginal model cost. Deterministic workflows. Private context that never leaves the machine.
Design-partner conversations open.
Voice request to an @openhome speaker. A local AI books a real Delta flight end to end on a fresh Minisforum X1 iGPU. No human clicks. No cloud agent. 73 seconds. The video is the demo.
Numbers below. 🧵
The constraint shapes the product. One founder. Sovereign infrastructure. Three domains.
If the system multiplies capability enough, team size stops mattering.
The bottleneck is not smarter agents or better models. It is the layer between them.
One person hits a ceiling not of talent but of coordination. The solution is structural, not more effort.
Correctness is not optional in trust infrastructure. The safety verifier, the identity system, and the knowledge engine cannot afford undefined behavior.