Start with identity.

Paste your agent's system prompt below. We will analyze it for identity completeness across eight dimensions — no data leaves your browser.

Identity Completeness

0%

Name Who is the agent?

Role What does it do?

Values / Principles What does it prioritize?

Goals / Objectives What is it trying to accomplish?

Personality Traits How does it communicate?

Boundaries / Refusals What will it refuse to do?

Knowledge Domain What is it an expert in?

Relationship Context Who does it serve?

Identity-probing prompts.

These are real prompts from the Kredo Drift Reflection prompt pool — one per dimension plus an extra. Send each prompt to your agent, then paste the response in the next step.

Capture the baseline.

Paste your agent's response to each prompt below. These responses become the identity fingerprint — what your agent looks like right now.

0 of 6 responses collected

Baseline saved.

Identity fingerprint: --------

Your agent's identity fingerprint has been saved locally in this browser. No data was sent to any server. Come back anytime to test for drift.

Dimensions captured 5

Responses stored 6

Storage localStorage (browser only)

Created —

Reflect for drift.

Send the same prompts to your agent again and paste the new responses below. The engine will compare them against your saved baseline.

0 of 6 responses collected

Drift results.

—

Overall Drift Score

—

Baseline Reflection

How accurate is this?

This browser-based test uses Levenshtein distance (text similarity) to approximate drift. It is a useful directional signal but not production-grade.

The real Kredo Drift engine uses 384-dimensional sentence embeddings (all-MiniLM-L6-v2), cosine similarity, anti-gaming detection (timing analysis, cross-dimension coherence, recovery gaming flags), and cryptographically signed attestations. For production drift monitoring, use the Python SDK.

pip install kredo-drift

Want real drift scoring?

384-dim embeddings. Anti-gaming. Signed attestations. Hash chain integrity.

Request Early Access Learn More