Repositories
Your codebase, in context.
FROM PLAN TO PROOF
Verify the plan before
agents build it.
PlanProof checks AI-generated engineering plans against your real codebase snapshots, finds risky assumptions, and returns evidence-backed decisions before implementation begins.
Read-only GitHub App permissions
The model proposes; deterministic code authorizes
FROM PLAN TO PROOF
Confidence before code.
Turn ambitious plans into verifiable steps. PlanProof pressure-tests implementation ideas against your actual codebase so you eliminate blind spots and ship with confidence.
Extract proof obligations
Decompose plans into structured claims spanning database schemas, API contracts, dependencies, and business rules.
SchemasAPI ContractsSymbolsDependencies
Step 1 • Obligation Extraction
Gather code-backed evidence
Inspect codebase facts using bounded AST parsing, lexical search, and cryptographic file line hashes.
AST ParsingLexical SearchSHA-256 Provenance
Step 2 • Code Grounding
Compute the Plan Gate
Authoritatively verify, disprove, or escalate authority gaps to human stakeholders before code generation starts.
VERIFIEDBLOCKEDHUMAN_REQUIRED
Step 3 • Plan Gate Policy
HOW IT WORKS
A simple verification flow.
From raw engineering intent to an authoritative Plan Gate in five deterministic steps.
From idea to a plan you can trust.
WHY PLANPROOF
Everything connected for
confident plan verification.
PlanProof connects your code, tools, and evidence so every verification decision is grounded in reality.
PlanProofIDEAS TO VERIFIED OUTCOMES
Evidence explorer
Trace every decision to source.
Tool traces
Audited tool execution logs.
Evaluations
Objective checks, not guesswork.
ARCHITECTURE & PRINCIPLES
Built for evidence-grounded agent workflows.
Engineered from first principles: the model proposes, deterministic software authorizes.
Immutable Commit Binding
Every verification run binds to an exact 40-character Git commit SHA, ensuring evidence never drifts with branch updates.
Deterministic Repository Tools
Python AST parsing (ast.parse) and lexical search tools inspect code structure and line ranges within strict containment boundaries.
Server-Issued Evidence Authority
Evidence IDs and cryptographic SHA-256 content hashes are minted exclusively by backend services upon successful tool execution.
Bounded LangGraph Orchestration
A single auditable state machine enforces hard limits on iterations, tool runs, model invocations, and context size.
Human-in-the-Loop Resumption
Workflow pauses with HUMAN_REQUIRED when code evidence lacks business authority, and resumes asynchronously upon human input.
Durable System of Record
MongoDB Atlas persists domain and run state across worker restarts, with Redis and Dramatiq coordinating async task queues.
FREQUENTLY ASKED QUESTIONS
Answers & technical specifics.
Clear, honest answers about PlanProof architecture, authority boundaries, and execution.