CLI tooling for
code-aware agents.

Weaver connects AI agents to a real codebase through concrete observe, act, and verify operations. The CLI stays thin, the daemon keeps context loaded, and Double-Lock plus Birdcage constrain the risky parts.

$ cargo install weaver

Quick Start

FIG 1.0: ARCHITECTURE

Illustrated loom fused with a software architecture diagram, showing indigo and vermilion threads moving through a daemon-like central hub with blueprint measurements and wave motifs.

SCALE: 1:1

Composable

Use one surface for structural queries, patch application, and diagnostics. The daemon streams JSONL to the CLI, which then forwards raw payloads or human-readable output as needed.

Explore Fusion

Safe

Block bad edits before they land. Double-Lock checks the change; Birdcage isolates untrusted tooling with seccomp-bpf and namespaces.

Security Model

Fast

Keep state in a resident daemon instead of re-parsing on every call. Rust binaries stream JSONL over Unix domain sockets for low-friction automation.

Daemon Specs

Interface

Command Reference

View all commands ->

weaver-cli — zsh — 80x24

➜ ~ weaver observe get-definition --output json --uri file:///src/main.rs --position 10:5

[{"uri":"file:///src/lib.rs","line":42,"column":17}]

➜ ~ weaver act apply-patch --output json < fix.diff

{"applied":true,"files":["src/lib.rs"]}

➜ ~ weaver verify diagnostics --output json --uri file:///src/lib.rs

View Full Documentation

QUERY ENGINE

Illustrated left-to-right Sempai query pipeline showing code entering a circular parsing chamber and flowing as structured output toward an agent target.

INPUT: RAW SOURCE OUTPUT: STRUCTURED JSONL

Query layer

Sempai Query Engine

Sempai adds Semgrep-compatible structural queries to observe. It runs on Tree-sitter, exposes a query-focused operator surface, and keeps compatibility boundaries explicit.