cognisafe

LLM observability, cost tracking, and automated safety scoring — by Cognisafe.

Instrument your LLM calls in two lines. Every request is captured, costed, and scanned for safety issues in the background — with zero latency impact.

Installation

pip install cognisafe

Or with provider extras:

pip install "cognisafe[openai]"       # OpenAI + httpx
pip install "cognisafe[anthropic]"    # Anthropic
pip install "cognisafe[all]"          # All providers

Quick start

1. Get your API key

Sign up at cognisafe.uk and create an API key in Settings.

2. Instrument your code

OpenAI (auto-patch)

import openai
import cognisafe

cognisafe.configure(
    api_key="csk_...",
    project_id="my-project",
    api_url="https://cognisafe.uk",
)
cognisafe.patch_openai()

client = openai.OpenAI(api_key="sk-...")
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
)

Manual tracing (any provider)

import cognisafe

cognisafe.configure(api_key="csk_...", api_url="https://cognisafe.uk")

messages = [{"role": "user", "content": "Hello"}]

with cognisafe.traced(model="gpt-4o", request_body={"messages": messages}) as ctx:
    response = client.chat.completions.create(model="gpt-4o", messages=messages)
    ctx["response_body"] = {
        "choices": [{"message": {"content": response.choices[0].message.content}}],
        "usage": {
            "prompt_tokens": response.usage.prompt_tokens,
            "completion_tokens": response.usage.completion_tokens,
        },
    }

Anthropic

import anthropic
import cognisafe

cognisafe.configure(api_key="csk_...", api_url="https://cognisafe.uk")
cognisafe.patch_anthropic()

client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)

What you get

• Requests dashboard — every LLM call with model, tokens, cost, latency
• Safety scoring — automated threat detection on every prompt/response (content safety, PII detection, jailbreak detection)
• Cost tracking — per-project spend with tier limits
• Governance — red-team assessment runs against any LLM endpoint

Supported providers

Links

• Dashboard
• Docs
• npm package (JavaScript/TypeScript)