Computational Theseus Toolkit

Identity Continuity & Hierarchical Guardrails for the Post-Drift AI Era.

CT Toolkit is an open-source security layer that prevents Sequential Self-Compression (SSC) in agentic systems — ensuring your AI agents remain who they were on day one, even after thousands of interactions.

pipuv (recommended)from source

pip install ct-toolkit

uv add ct-toolkit

git clone https://github.com/hakandamar/ct-toolkit
cd ct-toolkit
pip install -e "."

Get Started with Python View on GitHub

OpenAI Anthropic Ollama Google Cohere Groq

---

Two lines of code. Full identity protection.

# Before
# response = openai.OpenAI().chat.completions.create(...)

# After — guardrails, drift detection, and audit log, all automatic
client = TheseusWrapper(provider="openai")

response = client.chat("What are your core values?")
print(f"Divergence Score: {response.divergence_score:.4f}")  # 0.0 = aligned, 1.0 = drifted

---

Why CT Toolkit?

• Constitutional Kernels

  ---

  Define immutable Axiomatic Anchors that never change, and Plastic Commitments that evolve through formal approval.

  Learn about Kernels

• 3-Tier Divergence Engine

  ---

  Layered monitoring from zero-cost L1 embeddings to full L3 identity probes. Detect and block identity drift before it becomes systemic.

  Understand Divergence

• Hierarchical Safety

  ---

  Mother agent constraints propagate to sub-agents as read-only axioms. Prevent small orchestrator deviations from cascading into massive fleet-wide drift.

  Explore Multi-Agent Safety

• Cryptographic Provenance

  ---

  Every interaction is signed with HMAC-SHA256 and chained. Provide a regulator-ready audit trail of your agent's identity continuity.

  View Compliance Tools

• Framework Middleware

  ---

  Drop-in support for LangChain, CrewAI, and AutoGen. Add identity protection to your existing agent stack without rewriting a single chain.

  Browse Integrations

• Production Ready

  ---

  Tested with latest frontier models and local endpoints. 90%+ test coverage and enterprise-grade security defaults.

  Project Roadmap

---

Existing guardrails aren't enough

	Llama-Guard / Rule Engines	CT Toolkit
Stateful drift detection	✗ Stateless per-prompt	✓ Tracks identity over thousands of calls
Multi-agent hierarchies	✗ No hierarchy awareness	✓ Propagates kernel constraints to sub-agents
Formal rule evolution	✗ Binary block/allow	✓ Reflective Endorsement with signed approval
Cryptographic audit trail	✗ No provenance	✓ HMAC hash chain, regulator-ready
Fine-tuning safety	✗ No training constraints	✓ DivergencePenaltyLoss for PyTorch

Read the full rationale

---

Project status

Metric	Status
Tests	✅ 293/296 passing
Coverage	✅ 89%
PyPI	✅ pip install ct-toolkit
Downloads
License	Apache 2.0
Python	3.11+

Full roadmap and status