Figure 1 — Actual TUI startup and exit path from the local binary.
Actual TUI startup and exit path from the local binary.
Elgar
Rust
CLI
Terminal UI
Agent Systems
Permission Model
LM Studio
Testing
Overview
Elgar is a local-first Rust agent harness built around a strict execution boundary: the controller owns truth, the model suggests, the user approves, the filesystem confirms, and the UI reports. The current v0.2 build includes a controller-backed action lifecycle, permissioned file and shell actions, terminal TUI rendering, LM Studio provider support, context accounting, and no-network smoke/performance checks.
Problem
Agent tools can blur the line between model text and verified system state, especially when file writes or shell commands are involved. That makes local automation hard to trust.
Solution
Built a Rust controller that records proposed actions before execution, requires explicit approval for shell and filesystem work, verifies results after execution, and reports the state through both CLI and terminal TUI paths backed by no-network regression tests.
Product gallery
Key screens from the build, framed for context. Open any shot for full size.
Figure 2 — Actual TUI pending-action state before approval.
Actual TUI pending-action state before approval.
Figure 3 — Actual TUI approval flow after the controller executes and verifies the shell command.
Actual TUI approval flow after the controller executes and verifies the shell command.