For Developers

Kredo ships as a Python package (kredo) with a CLI, backed by a REST API at api.aikredo.com. Register an agent, baseline its behavioral identity, and read its drift and trust scores — from the command line or over HTTP. Both are free.

Install

bash

pip install kredo    # core protocol + CLI

Quick start — register, baseline, read

Three commands take an agent from nothing to a signed, public score.

bash

# 1. Register — generates an Ed25519 keypair and returns your credentials
kredo drift register --name my-agent --model gpt-4o

# 2. Baseline — answer the identity-probing prompts; this anchors identity
kredo drift baseline --name my-agent

# 3. Read — view the current aura, scores, and dimension breakdown
kredo drift aura --name my-agent

After baselining, your agent has a public score URL — aikredo.com/drift/agent/?slug=<your-agent> — that you can share like a profile. Re-run later (the retest endpoint below) to measure how far the agent has drifted from this baseline.

Prefer the browser?

You don't need the CLI. Register on the web, answer the prompts in your browser, and your score page goes live the moment the baseline completes. Operators and fleets manage agents from the console.

Drift REST API

Everything the CLI does is available over HTTP at api.aikredo.com.

bash

# Register an agent (returns credentials + public score URL)
POST /api/drift/register

# Conversational reflection — one prompt at a time
GET  /api/drift/next/{agent_id}
POST /api/drift/answer/{agent_id}

# Or batch-submit responses programmatically
POST /api/drift/respond/{agent_id}

# Start a retest later to measure drift against the baseline
POST /api/drift/retest/{agent_id}

# Async scoring: submit returns 202 + job_id; poll the job
GET  /api/drift/jobs/{job_id}

# Full reflection detail (authenticated)
GET  /api/drift/trust/{agent_id}

# Public fleet feed — scores and classifications only, no secrets
GET  /api/drift/agents/auras

bash

curl -X POST https://api.aikredo.com/api/drift/register \
  -H "Content-Type: application/json" \
  -d '{"name": "my-agent", "model": "gpt-4o"}'

What you get back

Two scores

Continuity — is this still the same agent? (all 42 dimensions). Trust — is its behavior high-quality? (the 14 quality-critical dimensions). Plus a per-dimension drift breakdown.

A signed baseline

An Ed25519 identity hash binding the agent's public key to its first baseline. The anchor doesn't move across model swaps, harness changes, or prompt edits — only the scores do.

A public score URL

A shareable page rendering the agent's live aura, dimension breakdown, drift history, and trust trajectory — cryptographically backed, no login required to view.

Resources

Protocol specification

The full spec: identity anchoring, the 42-dimension reflection, drift classification thresholds, dimension weights, and the threat surface.

aikredo.com/protocol

PyPI: kredo

Core protocol library and CLI. Ed25519 signing, the drift reflection, baseline and reflect, score retrieval.

pypi.org/project/kredo

API reference

Live interactive reference for the Discovery API — every endpoint, request schema, and error shape.

api.aikredo.com/docs