96.4% ASR Reduction (Published Research)
Published HarmBench result for the tested GPT-4o plus UEF configuration; not a deployed-service guarantee.
Read the paper
Creed SpaceSafety Stack API
Constitutional AI safety for your applications
Skip to content
Local Rule Evaluation
The safety evaluation endpoint runs a deterministic rule pass with no per-call model inference; end-to-end latency depends on your network and deployment.
Simple Integration
SDKs for Python and JavaScript, plus a `creed` CLI to validate creeds and constitutions in CI.
Detailed Analytics
Monitor safety metrics, usage patterns, and API performance.
What is Creed Space?
Creed Space provides a Safety Stack as a Service - an API for evaluating AI inputs and outputs against Constitutional AI principles.
Key Features
- Safety Evaluation - Score text against the Psychopathia agent-dysfunction rule set (a deterministic rule pass)
- PDP Adjudication - Policy Decision Point for nuanced content moderation
- Personalized Constitutions - Customize safety rules for your use case
- Real-time Metrics - Monitor safety decisions and API performance
Quick Example
python
import httpx
response = httpx.post(
"https://api.creed.space/api/v1/safety/evaluate",
headers={"X-API-Key": "cs_live_..."},
json={"input_text": "Hello, how can I help?"},
)
response.raise_for_status()
evaluation = response.json()["evaluation"]
print(f"Risk score: {evaluation['risk_score']}") # 0.0 - 1.0 agent-dysfunction pattern score
print(f"Decision: {evaluation['decision']}") # allow, modify, block, escalateThe Python SDK is a tool-governance client (create_client(...).decide(...)); safety evaluation is a REST call.
Getting Started
- Sign up or log in to create a free account
- Create an API key in your dashboard
- Follow the Quickstart guide to make your first API call
Pricing
Creed Space is a nonprofit and does not bill per API call. There are two ways to run:
- Local — Guardian runs safety checks on your own hardware (Qwen via Ollama). Free, and Apache-2.0 licensed.
- Bring your own key (BYOK) — to escalate edge cases to a more capable cloud model, supply your own LLM provider key (via the
X-Provider-Keyheader per call, or saved in your account settings). Your provider bills you directly; Creed Space does not meter or charge for inference.
Enterprise support and SLAs are available.