RiftPoint 🌌

High-performance multiversal checkpointer and speculative state engine for LangGraph, powered by Janus-Tachyon-RS.

🌟 Overview

In standard LLM agent workflows, exploring alternative reasoning paths, evaluating multiple tool-use strategies, or testing speculative futures requires expensive state copying or complex DAG orchestration.

RiftPoint bridges LangGraph with the Rust-backed janus-tachyon-rs engine to provide:

• Multiversal Branching: Instant state forks into candidate realities with near-zero memory duplication penalty.
• Speculative Tool Calling: Concurrent execution of divergent tool-use strategies across distinct timeline branches.
• Multiverse Collapse: Automatic scoring of candidate timelines (via LLM-as-a-judge, heuristics, or consensus) to commit winning trajectories to main while pruning discarded paths.
• Full State Travel & Tracing: Interactive time travel, undo/redo, and timeline visualization via Mermaid.js and Matplotlib.

📚 Planning & Architecture Documentation

Detailed architectural blueprints and implementation phases are documented in docs/planning:

• Master Architecture & Overview Plan
• Performance Benchmarks & Multiverse Capability Report
• Future Directions & Strategic Roadmap
• Phase 1: Core Janus Bridge & Sync Checkpointer
• Phase 2: Asynchronous Checkpointer
• Phase 3: Branching & Speculative Execution Engine
• Phase 4: Multiverse Collapse & Evaluation Engine
• Phase 5: Visualization, Serialization & Tracing

🚀 Quickstart

Installation

pip install riftpoint

Or install with uv:

uv add riftpoint

Basic Example

from langgraph.graph import StateGraph
from riftpoint import RiftCheckpointSaver

# Initialize the Janus-backed checkpointer
checkpointer = RiftCheckpointSaver()

# Build your LangGraph workflow
builder = StateGraph(...)
# ... define nodes and edges ...
graph = builder.compile(checkpointer=checkpointer)

# Run workflow with multiversal checkpointing
config = {"configurable": {"thread_id": "session-1"}}
result = graph.invoke({"messages": [...]}, config=config)

# Render the timeline DAG as a Mermaid diagram
print(checkpointer.visualize("session-1"))

🛠️ Developer Commands

Development workflows use uv:

# Sync dependencies
uv sync --all-groups

# Run tests with coverage requirement (>= 70%)
uv run pytest

# Lint and format
uv run ruff check --fix
uv run ruff format

# Type check
uv run mypy src tests

# Code complexity check (Radon Grade B or better)
uv run python scripts/check_radon.py src

📚 Documentation & Roadmap

• Documentation Site: Full guides, API reference, and tutorials.
• Roadmap to v1.0: 5-pillar production readiness checklist toward GA release.
• Future Directions: Long-term vision and feature exploration horizon.

📄 License

This project is dual-licensed under:

• MIT License (LICENSE-MIT)
• Apache License, Version 2.0 (LICENSE-APACHE)