ROADMAP.md

October 16, 2025 ยท View on GitHub

Feature Roadmap

This document outlines the roadmap for the Binharic coding agent. It includes planned features, improvements, and their current implementation status.

Important

This roadmap is a work in progress and is subject to change without notice.

1. Core Agentic Capabilities

  • LLM Provider Support
    • OpenAI models (GPT-4o, GPT-4o-mini, and GPT-5 series)
    • Anthropic models (Claude 4 Sonnet and Claude 4.5 Sonnet)
    • Google AI models (Gemini 2.5 Pro and Gemini 2.5 Flash)
    • Ollama for local model execution
    • Support for Azure OpenAI endpoints
    • Support for additional model providers (Cohere and Mistral)
  • Context Management
    • Token-based context window management
    • Automatic context trimming for long conversations
    • History preservation across sessions
    • Adaptive context management with prepareStep callbacks
    • Tool result summarization for longer agentic loops
    • Intelligent context compression
    • Semantic context pruning
  • Multi-Step Execution
    • Multi-step tool calling with retry logic
    • Transient error handling with exponential backoff
    • Tool execution confirmation flow
    • AI SDK 5 Agent class integration (currently using AI SDK streaming plus tool calling directly)
    • Automatic loop control with stopWhen conditions
    • Budget-based stopping conditions
    • Error threshold conditions
    • Validation-based stopping
    • Completion detection conditions
    • Parallel tool execution for independent operations
    • Automatic tool dependency resolution
  • Specialized Agents
    • Main Binharic Agent (Tech-Priest persona)
    • Code Analysis Agent
    • Security Audit Agent
    • Test Generation Agent
    • Documentation Agent
    • Refactoring Agent
    • Debug Agent
    • Performance Optimization Agent
    • Migration Agent

2. Tool System

  • File Operations
    • Read files with tracking
    • Create new files
    • Smart file editing with diff application
    • List directory contents
    • File search capabilities
    • Integrated with AI SDK tool calling
    • Automatic file tracking for edits (no manual read required)
    • File staleness detection
    • Memory-efficient file tracking (1000 file limit)
    • File comparison and diffing
    • Bulk file operations
    • File watching for changes
  • Code Intelligence
    • TypeScript/JavaScript error detection
    • Syntax validation
    • AI-powered code analysis
    • Structured refactoring suggestion
    • Automatic test generation
    • Security vulnerability detection
    • Code navigation (go to definition, find references)
    • Code formatting integration
    • Linting integration
  • Shell Integration
    • Bash command execution
    • Terminal session management with persistent state
    • Command timeout handling
    • Interactive shell support
    • Command history and replay
    • Environment variable management
  • Web and Network
    • URL fetching with HTML-to-text conversion
    • Content markup stripping
    • API integration templates
    • Web scraping capabilities
    • Webhook support
  • Model Context Protocol (MCP)
    • MCP server integration
    • Dynamic tool discovery
    • Stdio transport support
    • HTTP transport support
    • MCP tool caching
    • Custom MCP server templates

3. User Interface

  • Terminal UI (TUI)
    • Rich terminal interface with Ink
    • Real-time streaming responses
    • Tool execution confirmation prompts
    • Command history navigation (up/down arrows)
    • Help menu
    • File search with @ mention
    • Non-blocking UI during LLM responses
    • Command syntax highlighting (partial match in yellow, full match in cyan)
    • Colored help menu items**
    • Clean message display (no "Binharic:" prefix)
    • Dynamic username from system (not hardcoded)
    • Tool results hidden from UI (only failures shown)
    • Git branch display in footer
    • Syntax highlighting for code blocks
    • Multi-pane view for side-by-side comparison
    • Terminal themes support
  • User Experience
    • Graceful error handling
    • Status indicators (idle, responding, tool-request, executing-tool, and error)
    • Git branch display
    • Responsive input field (non-blocking)
    • Clear error messages for tool failures
    • Exit summary screen on quit (session ID, tool calls, success rate, timings, model usage)
    • Progress bars for long operations
    • Notification system
    • Undo/redo for file operations
    • Session saving and loading

4. Configuration and Customization

  • Configuration Management
    • JSON5 configuration format
    • Model configuration with context windows
    • Custom system prompts
    • API key management
    • History size limits
    • Configuration profiles (development, production, etc.)
    • Configuration validation with detailed error messages
    • Hot-reload configuration changes
  • Personality and Behavior
    • Adeptus Mechanicus character and terminology
    • Customizable system prompts
    • Multiple personality presets
    • Conversation style customization
    • Output verbosity levels

5. Performance and Reliability

  • Error Handling
    • Categorized errors (Fatal, Transient, Tool)
    • Automatic retry with backoff for transient errors
    • Error logging with Winston
    • Graceful degradation
    • Agent-level error handling
    • Tool call ID tracking and validation (AI SDK 5)
    • History rollback on errors
    • Stream timeout protection (2 minutes)
    • Tool execution timeout protection (10 seconds for autofix)
    • Error recovery suggestions
    • Automatic error reporting (opt-in)
    • Configurable stderr suppression via env flag (planned)
  • Optimization
    • Efficient token counting
    • Context window optimization
    • Agent-based loop optimization
    • Specialized agents for specific tasks
    • Proper AI SDK 5 streamText result handling
    • Tool arguments extraction for both static and dynamic tools
    • Response caching
    • Request batching
    • Streaming optimizations
  • Monitoring
    • Structured logging
    • Provider availability checks
    • Detailed tool execution logging
    • Autofix attempt tracking
    • Basic session metrics rendered on exit (LLM API time, tool time, request counts)
    • Persistent performance metrics collection
    • Usage analytics (tokens, costs)
    • Health checks and diagnostics

6. Testing and Quality

  • Test Coverage
    • Unit tests for core functionality
    • Tool execution tests
    • Error handling tests
    • State management tests
    • UI component tests
    • Comprehensive tool call ID mismatch and history rollback tests
    • Extensive overall test suite (hundreds of passing tests)
    • Integration tests
    • End-to-end tests
    • Performance benchmarks
  • Code Quality
    • TypeScript strict mode
    • ESLint configuration
    • Prettier formatting
    • Pre-commit hooks
    • Comprehensive error typing
    • Automated dependency updates
    • Security scanning
    • Code complexity metrics

7. Documentation

  • User Documentation
    • README with basic usage
    • API key setup instructions
    • Contributing guidelines
    • Comprehensive user guide
    • Video tutorials
    • FAQ section
    • Docker/Container usage guide (planned)
  • Developer Documentation
    • Code of conduct
    • Architecture documentation
    • Bug fix documentation
    • API reference documentation
    • Plugin development guide
    • Deployment guide

8. Distribution and Deployment

  • Package Management
    • NPM package structure
    • TypeScript compilation
    • NPM registry publication
    • Semantic versioning (via git tags)
    • Release automation (GitHub Actions: npm + GHCR)
  • Installation Methods
    • Homebrew formula (macOS)
    • Snap package (Linux)
    • Chocolatey package (Windows)
    • Docker image
      • Published to GitHub Container Registry: ghcr.io/<owner>/<repo>
      • Multi-arch builds (linux/amd64, linux/arm64) via Buildx
      • Makefile targets for local and CI builds/pushes
      • Optimized build context via comprehensive .dockerignore
    • Standalone binary releases
  • Cloud and Remote
    • Remote execution support
    • Multi-user deployments
    • Cloud provider integrations

9. Advanced Features

  • Agentic Capabilities
    • Autonomous task execution
    • Tool chaining
    • File tracking for safe edits
    • AI SDK 5 Agent class for reusable configurations
    • Multi-step tool execution with automatic loop control
    • Specialized agents with distinct personalities
    • onStepFinish callbacks for monitoring
    • prepareStep callbacks for dynamic configuration**
    • Multiple stopping conditions (step count, budget, errors, validation, completion)
    • Goal-oriented planning
    • Task decomposition
    • Long-term memory
    • Agent composition (combining multiple agents)
    • Dynamic agent selection
  • Collaboration
    • Team workspaces
    • Shared conversation history
    • Code review assistance
    • Pull request analysis
    • Multi-agent collaboration
  • Extensions and Plugins
    • Plugin system architecture
    • Custom tool registration
    • Language-specific plugins
    • Framework-specific assistants
    • Custom agent templates

10. Security and Privacy

  • Security
    • API key environment variable support
    • Secure credential storage
    • Encrypted configuration files
    • Sandboxed tool execution
    • Rate limiting
  • Privacy
    • Local execution option (Ollama)
    • Conversation history stored locally
    • Data anonymization options
    • GDPR compliance
    • Audit logging