Runtime-Owned
Coding Agent

The runtime, not the model, owns routing, tool policy, verification, and completion. Local-first CLI agent built in Rust.

GitHub Releases Architecture ↓

Runtime-Owned Orchestration

Model answers are candidates, not completions. The runtime classifies intent, builds execution strategy, enforces tool policy, runs verification, and adjudicates completion through a 5-gate lattice.

PostgreSQL 18 + pgvector unifies control-plane, retrieval, and conversation memory in one operational backend.

builder

// user input received

classify IntentClassifier → Task

[route] ExecutionStrategy: Task

[gate] 5 acceptance gates opened

phase Discover → Plan → Execute

[tool] policy-gated execution ✓

[verify] completion verification run

proof building gate evidence…

[adj] adjudicate → Complete ✓

TASK COMPLETED

Runtime Verification

5-gate completion lattice with structured evidence, not model self-assessment.

Weak-Model Hardening

SelfReview gates, citation breadth requirements, and escalation to standard agents.

Hybrid Retrieval

Symbol-level packetization + deterministic embeddings + pgvector hybrid search.

Execution Strategy Matrix

Strategy	Tool Classes	Completion Contract
Task	read+search+write+shell+delegate	citation≥1, no broad claim
Survey	read+search+delegate	citation≥3, no broad claim
Review	read+search only	citation≥2, no broad claim
Delegate	read+search+delegate	citation≥1
Conversation	none	permissive

Getting Started

Quick Start

01 Install globally

terminal
$ npm install -g builder

02 Run interactively

builder
$ cargo run --bin builder -- "Summarize the
                                current workspace"

Runtime-Owned Coding Agent

Runtime-Owned
Coding Agent