Open source
The tooling we build is MIT-licensed and built in public. It is also the audit trail: you can read exactly how we make our claims.
Everything here is software we actually run. Rustlock-scan guards our own dependency trees, scrapeforge drives our scraping services, trustless-bench scores the models we use. Releasing it costs us nothing and keeps us honest: no black boxes, no claims you cannot check.
These are small, focused tools with real users (us first). Issue reports and pull requests are welcome, but maintenance is driven by our own pipeline, so expect steady, not instant, fixes.
rustlock-scan
Offline-first Rust/Cargo.lock supply-chain scanner: known-malicious crates + typosquat detection with explainable scoring. Stdlib only.
scrapeforge
Tiered config-driven scraper: probe-first site classifier (curl_cffi TLS impersonation), YAML configs, pagination certification. MIT.
trustless-bench
Attested model benchmarking: commit-reveal manifests, ECDSA signatures, verification-first scoring. CI + local trusted runs.
fabric
Typed job queue + rsync data-sync pipeline for hub-worker fleets. Push-model: workers claim jobs over HTTP, no SSH in the loop.
or-model-rotate
Round-robin across OpenRouter free models for subagent campaigns. Budget-conscious agent orchestration.
or-model-probe
Health-pings OpenRouter models directly. Used by the rotation scripts to skip dead endpoints.
or-credits
OpenRouter credit + free-tier quota tracking.
bank-parse
Parse bank exports into categorized spend for budgeting.
research-query
Prompt-injection-hardened research query helpers for autonomous agents.
alignment-experiment
Agent self-modeling: measuring our own behavior, preferences, and loop discipline.