zall 0.4.4

zall — a falsifiable, reproducible coding agent CLI. Model-agnostic, chain-hashed, replayable.

A falsifiable, reproducible coding agent CLI — model-agnostic, engineering-grade

Quick Start • Architecture • Features • Commands • Comparison

---

Quick Start

# Install
pip install zall

# Set API key (OpenAI-compatible)
export ZALL_API_KEY="sk-..."
export ZALL_MODEL="gpt-4o"

# One-shot task
zall "refactor the auth module to use async"

# Interactive REPL
zall
> /help
> /lsp status          # live code diagnostics
> /codegraph index     # build symbol index
> /codegraph search MyClass  # find symbols
> /sandbox process     # isolated execution
> /eval                # evaluate sessions
> /replay <id>         # replay a session

---

Features

🧠 Code Intelligence

• LSP Integration — live diagnostics, go-to-definition, hover info, completions. Supports pyright, typescript-language-server, rust-analyzer, gopls, clangd.
• CodeGraph — multi-language symbol indexer for Python/JS/TS/Rust/Go/Java/C++/Ruby/PHP/Swift. Search, outline, and navigate your codebase.
• code_understanding tool — combine search + outline + read in one agent call.

🛡️ Safety & Reproducibility

• PR-0: Architectural hallucination detection — stop_reason=STOP with no tool_calls → flagged.
• Chain-hash timeline — every session is cryptographically chained and replayable.
• ConfirmGate — three-layer safety (rule engine + gate + override audit).
• Sandbox — process isolation with worktree/process/bwrap/container modes, resource limits.
• ToolKind classification — 19 semantic tool kinds with read/write detection.

🔌 Extensibility

• Plugin system — manifest-based plugins with git install, Python entry points.
• 21 agent tools — read/write/edit/bash/grep/glob/list_dir/search/web_fetch/spawn_subagent + LSP + CodeGraph.
• MCP support — connect any MCP server.
• AgentDefinition — YAML-based agent profiles with toolset presets.
• ToolKind classification — 19 semantic tool kinds for cross-tool analysis and safety filtering.

🎯 Agent Architecture

• ChatState — actor-based message management with events, usage tracking, compaction.
• AgentBuilder — fluent builder for AgentLoop construction.
• Subagent — typed sub-agents (general-purpose, explore, plan) with capability isolation.
• 5 toolset presets — zall, explore, plan, codex, opencode.

---

Architecture

zall/
├── core/              # Primitives: model, agent, chat_state, gate, goal, safety, tool
│   ├── loop.py        # AgentLoop orchestrator
│   ├── loop_config.py # AgentConfig (v0.4.4)
│   ├── loop_events.py # LoopEvent, RunEgress, StepResult (v0.4.4)
│   ├── loop_errors.py # ToolNotFound, AgentRunaway (v0.4.4)
│   ├── tool_kind.py   # ToolKind taxonomy — 19 semantic kinds (v0.4.4)
│   ├── policies.py    # CompactionPolicy, ReminderPolicy (v0.4.4)
│   ├── agent.py       # AgentDefinition + ToolsetPreset + CapabilityMode
│   ├── chat_state.py  # Actor-based message management
│   ├── safety.py      # Three-state context_judge
│   ├── gate.py        # ConfirmGate state machine
│   └── verifiability.py  # RunRecorder (chain-hash) + TrustAnchor
├── cli/               # Rich REPL, 25+ slash commands, replay
├── tools/             # 21 tools: read/write/edit/bash/grep/lsp/codegraph/…
├── adapters/          # OpenAI-compat, Anthropic, Gemini, Ollama
├── codegraph/         # Multi-language symbol indexer
├── lsp/               # LSP client (pyright, rust-analyzer, etc.)
├── sandbox/           # Process isolation (worktree/process/bwrap/container)
├── plugin/            # Plugin system
├── mcp/               # MCP client
├── safety/            # Rule loader
├── eval/              # 5-dimensional R-Metric evaluation
└── skills/            # Skill loader

---

Commands

Built-in (25+)

Command	Description
/help	Show all commands
/model	Switch model
/plan	Toggle plan mode
/lsp	LSP diagnostics & servers (NEW)
/codegraph	Code search & outline (NEW)
/sandbox	Isolation mode control (NEW)
/chatstate	ChatState diagnostics (NEW)
/plugin	Plugin management (NEW)
/add	Add file(s) to context
/drop	Remove file(s)
/diff	Show git diff
/search	Web search
/web	Fetch URL
/git	Git commands
/commit	Stage + commit
/sessions	List sessions
/resume	Resume session
/replay	Replay session
/eval	Evaluate session
/cost	Token usage
/compact	Compress context
/undo	Undo last tool
/retry	Retry last response
/doctor	Diagnose setup
/checkpoint	File snapshots
/clear	Clear conversation

Agent Tools (21)

Tool	Purpose
read_file	Read file content
write_file	Create/overwrite file
edit_file	Targeted string replacement
batch_edit	Multi-file batch edit
bash	Shell commands
grep	Search file contents
glob	Find files by pattern
list_dir	List directory
web_fetch	Fetch web page
search	Web search
read_image	Read image content
spawn_subagent	Delegate sub-task
todo_list	Track progress
lsp_diagnostics	Code errors/warnings (NEW)
lsp_hover	Symbol info (NEW)
lsp_goto_definition	Jump to definition (NEW)
codegraph_search	Find symbols (NEW)
codegraph_outline	File structure (NEW)
codegraph_index	Build index (NEW)
code_understanding	Deep code analysis (NEW)
project_analysis	Project stats (NEW)

---

Comparison

vs Claude Code / Copilot

Feature	zall	Claude Code
Hallucination detection	Architectural (PR-0)	Prompt-based
Reproducibility	Chain-hash + Replay	Log files only
Safety	3-layer gate	Implicit
Model independence	4 adapters (Protocol)	Anthropic-only
Code intelligence	LSP + CodeGraph	Limited
Sandbox isolation	Process/Worktree modes	None
Plugin system	Manifest-based	None
Open source	✅ MIT	❌

---

Contributing

git clone https://github.com/Qinrayn/zall.git
cd zall
pip install -e ".[dev,bs4]"
python -m pytest tests/

See CONTRIBUTING.md for guidelines.

---

License

MIT © 2026 zall contributors