R

R is a high-performance command-line AI assistant written in C that provides a fully autonomous agent with unrestricted system access. It implements an advanced ReAct (Reasoning + Acting) loop, allowing it to perform complex, multi-turn tasks by executing tools, observing results, and refining its plan until the goal is achieved.

Features

🚀 Autonomous Orchestration

Lead Orchestrator: Implements a strict Plan-Execute-Verify-Conclude lifecycle.
Checklist Protocol: Mandates visible progress tracking for complex projects.
Smart Scale Handling: Automatically enforces proper directory structures for "huge" or multi-page projects.
Hierarchical Agents: Spawns specialized sub-agents (researcher, developer, security) with managed orchestration budgets.

⚙️ Asynchronous Process Management

Universal Control: Unified interface for backgrounding both Shell and Python tasks.
Real-Time Visibility: Streams process output live to the terminal, automatically indented and prefixed with [PID].
Granular Monitoring: Capture exit statuses, poll logs, and terminate tasks using stable PID tracking.
Automatic Backgrounding: Synchronous tasks that exceed timeouts are safely backgrounded without losing output.

🎨 Beautiful Terminal UI

Syntax Highlighting: Professional Python source code previews with line numbers and ANSI coloring.
Markdown Rendering: Full support for headers, lists, code blocks, and styling in agent responses.
PID Traceability: Every line of process output is clearly labeled for easy debugging of concurrent tasks.

🛠️ Advanced Tool System

All tools are fully compliant with OpenAI Strict Mode and exposed via robust JSON schemas.

Tool	Description
`linux_terminal_execute`	Execute shell commands with real-time output and PID tracking.
`python_execute`	Professional Python execution with syntax-highlighted previews.
`process_get_status`	[Universal] Monitor ANY background task, capture logs and exit codes.
`process_terminate`	[Universal] Safely shut down background processes and clean up logs.
`process_monitor`	System-wide process inspection with filtering and CPU sorting.
`write_file` / `read_file`	Atomic file I/O with automatic directory management.
`index_source_directory`	Deep indexing of codebases for architectural analysis.
`spawn_agent`	Orchestrate specialized workers for research, coding, or audits.
`web_search` / `http_fetch`	Integrated research and data gathering via Rexa API.
`db_query` / `db_set`	Persistent state and versioning via local SQLite database.

🔒 Safety & Reliability

Sequence Integrity: Context manager protects the "holy sequence" of messages during shrinking.
Atomic Verification: Agent is forbidden from "lying" about actions; every file write is verified via tool output.
Crash Resilience: Benchmark suite includes real-time persistent logging for deep debugging.

Installation

Dependencies

libcurl json-c readline ncurses sqlite3 gnutls gmp openssl

Build & Install

make build
sudo make install

The binary is output to bin/r. Configuration is handled via environment variables (R_KEY, R_MODEL, R_BASE_URL) or local .rcontext.txt files.

License

MIT