Swades Agent

Swades Agent is an open-source, terminal-native autonomous AI coding agent built on the ReAct loop pattern. It embodies XERV's mission of Integrating Intelligence In Physical Intelligence by bringing seamless, self-correcting automation directly to your desktop environments.

swades-agent terminal preview
Fast-Forwarded Autonomous ReAct Execution
The Swades Agent Logo

ReAct Core Loop

Swades operates on a Reasoning + Acting loop. It processes tasks by observing, reasoning, calling tools, and iterating until the job is resolved.

24/7 Director Mode
A supervising Director model reviews progress after each run and writes the next subtask, iterating autonomously for continuous integration.
Real-time Token Streaming
Observe the agent's logic and tool arguments token-by-token directly in your terminal.

CUA Mode & Wayland Native Support

Swades introduces a graphical Computer Use Agent (CUA) mode for interacting directly with your desktop. It uniquely supports native GUI automation on modern GNOME Wayland configurations.

Mutter D-Bus Session Linking
Bypasses strict Wayland security restrictions using GNOME's native Mutter remote desktop infrastructure for pointer injection.
State-Based Mouse Tracking
Tracks absolute coordinates internally to draw target crosshairs via standard portal screencasting, without querying global pointers.
Anti-Loop Click Protection
Prevents getting stuck in infinite loops by strictly blocking consecutive clicks within a 25px x 15px bounding box.

Advanced Subagents System (v3.0)

For high complexity tasks, the parent orchestrator breaks down workflows and spawns parallel Subagents.

Git Worktree Isolation
Subagents execute in isolated workspace directories to prevent file corruption and dirty edits during development.
Automated Conflict Resolution
If modifications cause git merge conflicts, a specialized Merge-Resolution Subagent is spawned to safely resolve overlapping lines.

Sandbox Simulation Engine

Before committing changes to real-life files, the Simulation Engine tests alternative implementations in transient sandboxes.

Multi-Scenario Evaluation
Generates 2 to 4 alternative implementation scenarios, compiles them, and runs automated test suites.
Promotion Pipeline
The winning scenario proceeds through a strict verification pipeline: Workspace Git Rebase Check, Shadow Verification, and finally, Live Workspace Mutation.