claraity-code 0.13.0

ClarAIty Code - An AI-powered coding agent with TUI interface, tool execution, multi-layered memory, prompt caching, and intelligent context management

ClarAIty Code

An AI-powered coding agent with a professional TUI, 30+ built-in tools, subagent delegation, and multi-LLM support. Works with Claude, GPT, Qwen, DeepSeek, Ollama, and any OpenAI-compatible API.

Quick Start

pip install claraity-code

Set your LLM endpoint in .clarity/config.yaml (or environment variables):

export LLM_API_KEY=your-api-key
export LLM_BASE_URL=https://api.openai.com/v1
export LLM_MODEL=gpt-4o

Launch the TUI:

claraity --tui

Or use simple chat mode:

claraity

Core Features

Professional TUI Interface

Built on Textual with streaming responses, syntax-highlighted code blocks, collapsible tool-call cards, and real-time progress indicators. Falls back to a Rich-based CLI for headless environments.

30+ Built-in Tools

Category	Tools	Description
File Operations	read_file, write_file, edit_file, append_to_file, list_directory	Stream with line ranges, create, edit, append
Code Search	search_code, grep, glob, analyze_code, get_file_outline, get_symbol_context	Regex, glob patterns, AST parsing, LSP-based outlines
Git	git_status, git_diff, git_commit	Status, diffs, commits
Testing	run_tests, detect_test_framework	Auto-detect pytest/jest/vitest/cargo, fix suggestions
Web	web_search, web_fetch	Search the web, fetch and extract URL content
Execution	run_command	Shell command execution with approval gating
Planning	enter_plan_mode, request_plan_approval	Read-only planning with user sign-off
Delegation	delegate_to_subagent	Route tasks to specialist subagents
Other	create_checkpoint, clarify	Save progress, ask structured questions

Multi-LLM Support

Works with any OpenAI-compatible API. Tested with:

• Anthropic Claude (Sonnet, Opus, Haiku) -- with prompt caching
• OpenAI GPT (GPT-4o, GPT-4.1, o3)
• Alibaba Qwen (Qwen3-Coder, Qwen2.5-Coder)
• DeepSeek (DeepSeek-V3, DeepSeek-Coder)
• Ollama (any local model)
• Any OpenAI-compatible endpoint (vLLM, LM Studio, LocalAI, llama.cpp)

Prompt Caching

Automatic prompt caching for Anthropic and compatible providers. The built-in CacheTracker reports cache hit rates and effective cost savings (typically 50-80% reduction on long sessions). Cache metrics are logged per session.

Session Persistence

Every session is saved as append-only JSONL in .clarity/sessions/. Resume any previous session with full conversation history, tool call results, and context intact.

Structured Logging & Observability

All logs go to .clarity/logs/app.jsonl as structured JSONL -- no console noise, no TUI interference. Query logs with:

python -m src.observability.log_query --tail 50

Performance metrics and error tracking stored in .clarity/metrics.db (SQLite).

SubAgents

The agent delegates specialized tasks to focused subagents, each with their own system prompt and optional tool restrictions:

SubAgent	Purpose
code-reviewer	Analyzes code quality, security vulnerabilities, performance issues, and best practices
test-writer	Creates comprehensive test suites with unit, integration, and edge case coverage
doc-writer	Writes clear technical documentation for APIs, guides, and architecture
code-writer	Implements minimum code to satisfy requirements -- focused on writing, not exploration
explore	Fast read-only codebase explorer for finding code and tracing execution flows
planner	Produces detailed step-by-step implementation plans without writing code
general-purpose	Versatile agent with full tool access for multi-step research and implementation
knowledge-builder	Autonomously explores a codebase and generates structured knowledge base files

Each subagent can use a different LLM model (configured per-agent in config.yaml). Transcripts are saved to .clarity/sessions/subagents/.

Knowledge Base

ClarAIty loads project knowledge automatically from .clarity/knowledge/:

.clarity/
  knowledge/
    architecture.md      # System design overview
    conventions.md       # Coding standards
    api-patterns.md      # API design patterns
    deployment.md        # Deployment procedures

All .md files in this directory are loaded into the agent's context at startup. Use the knowledge-builder subagent to auto-generate these files from your codebase.

File-based memory hierarchy (4 levels, highest to lowest priority):

1. Enterprise -- /etc/clarity/memory.md (Linux) or C:/ProgramData/clarity/memory.md (Windows)
2. User -- ~/.clarity/memory.md
3. Project -- ./.clarity/memory.md (version-controlled, team-shareable)
4. Imports -- @path/to/file.md syntax for modular includes (circular detection built-in)

Permission Modes

Three modes control how the agent handles tool approval:

Mode	Behavior	Toggle
Normal (default)	Asks approval for write/edit/run operations	/mode n or Alt+M
Auto	Executes all tools without asking	/mode a
Plan	Read-only -- write operations are blocked entirely	/mode p

Switch modes mid-session with /mode or Alt+M in the TUI.

MCP Integration

Extend the agent with external services via Model Context Protocol servers. MCP tools are discovered dynamically and merged with native tools.

Configure in .clarity/config.yaml:

mcp_servers:
  - name: jira
    server_url: http://localhost:3000/sse
    tool_prefix: jira_
    connect_timeout: 30
    invoke_timeout: 60

Supports both SSE (remote) and stdio (local) transports. Tools are namespaced with configurable prefixes to avoid collisions.

Configuration

All configuration lives in .clarity/config.yaml:

logging:
  level: INFO
  handlers:
    jsonl: INFO
  retention:
    jsonl_max_bytes: 52428800

llm:
  backend_type: openai
  base_url: https://api.openai.com/v1
  model: gpt-4o
  context_window: 128000
  temperature: 0.2
  max_tokens: 16384

  # Per-subagent model overrides
  subagents:
    code-reviewer:
      model: claude-sonnet-4-5-20250929
    test-writer:
      model: claude-sonnet-4-5-20250929

Environment variables override config file values:

Variable	Purpose
LLM_API_KEY	API key for the LLM provider
LLM_BASE_URL	API endpoint URL
LLM_MODEL	Model name
LLM_BACKEND_TYPE	Backend type (openai, ollama)

Supported backends: openai (default), ollama, vllm, localai, llamacpp

Use Cases

Feature implementation -- Ask the agent to add a login page. It reads existing code, plans the approach, writes implementation files, and generates tests using the test-writer subagent.

Code review -- Point it at a PR diff or a set of files. The code-reviewer subagent checks for security issues, performance problems, and style violations with a verification-first methodology.

Codebase exploration -- Ask "how does authentication work?" The explore subagent traces execution flows across files and reports back with file paths and line numbers.

Test generation -- The test-writer subagent analyzes your code, detects the test framework (pytest, jest, vitest, cargo), and generates comprehensive test suites matching existing patterns.

Documentation -- The doc-writer subagent generates API docs, architecture guides, or inline documentation based on actual code analysis.

Installation

From PyPI

pip install claraity-code

From Source

git clone <repo-url>
cd ai-coding-agent
pip install -e ".[dev]"

Requires Python 3.10+.

LLM Setup

ClarAIty needs an LLM backend. Choose one:

Cloud API (Claude, GPT, etc.):

export LLM_API_KEY=your-key
export LLM_BASE_URL=https://api.openai.com/v1
export LLM_MODEL=gpt-4o

Local with Ollama:

# Install Ollama from https://ollama.ai
ollama pull qwen2.5-coder:7b
export LLM_BACKEND_TYPE=ollama
export LLM_MODEL=qwen2.5-coder:7b

Development

# Run tests
pytest tests/

# Run a specific test file
pytest tests/tools/test_file_operations.py -v

# Query structured logs
python -m src.observability.log_query --tail 50

# Launch TUI in development
python -m src.cli --tui

License

MIT License. See LICENSE for details.