Proofline
Proofline is a verification-first Google ADK agent designed to enforce evidence before action. It converts a work contract into explicit requirements, checks for fresh authoritative evidence, exposes contradictions, and creates a tamper-evident proof packet before an external action can proceed. The project uses Gemini 3.6 Flash on Google Cloud Run and applies a deterministic final gate on every run. The agent exposes distinct states such as NEEDS_EVIDENCE, CONFLICT, APPROVAL_REQUIRED, and READY. These states make missing evidence or conflicting sources visible rather than allowing fluent model output to be treated as completion. A live model call handles task decomposition and interpretation, while the deterministic gate ensures that unsupported claims cannot silently pass into action. Proofline provides a demo and source code, including an App discovery endpoint, OpenAPI documentation, and tests. The demonstration uses synthetic fixtures and shows a continuous run where evidence collection, a promise mismatch, repair checking, and human approval occur before the final packet is sealed. The proof packet includes a packet hash, and external action remains human-gated.
- Category
- AI Agents
- Pricing
- Free
Problem solved
AI agents can produce fluent output that appears to complete a task while lacking fresh, authoritative evidence or hiding contradictions. Proofline addresses this by converting task claims into explicit requirements, verifying evidence against authoritative sources, surfacing conflicts, and requiring a deterministic gate plus human approval before external actions are executed.
Use cases
- Verifying AI agent task claims before executing external actions
- Auditing whether requirements have fresh authoritative evidence
- Detecting contradictions across evidence sources in automated workflows
- Generating tamper-evident proof packets for compliance or handoff
- Gating deployment, financial, or operational actions with human approval
- Testing agent reliability with synthetic fixtures and deterministic gates
- Demonstrating verification-first agent design using Google ADK and Gemini
Audience
- Developers
- ML Engineers
- Enterprise
- Researchers
Pros
- Verification-first architecture with a deterministic final gate
- Exposes missing evidence and contradictions before action
- Human-in-the-loop approval for external actions
- Creates tamper-evident proof packets with a packet hash
- Source code, tests, and live agent endpoints are provided
Cons
- Requires access to fresh authoritative evidence sources
- Adds approval friction to automated workflows
- Demo relies on synthetic fixtures rather than production-scale deployments
- Depends on Google ADK, Gemini, and Cloud Run
- Limited public details about license, adoption, and release history
Official website: https://daradigu.com/proofline
Canonical LambdaBase page: https://www.lambdabase.com/tools/proofline