atomlabs 1.0.0

ATOM AI Governance SDK — pre-execution governance for every AI call

atomlabs

Pre-execution AI governance for every model you use.

Install

pip install git+https://github.com/qstackfield/atomlabs-sdk.git

Or for local development:

git clone https://github.com/qstackfield/atomlabs-sdk.git
cd atomlabs-sdk
pip install -e .

Requires Python 3.8+ and requests>=2.28.0.

Quick Start

from atomlabs import ATOMClient, GovernanceBlockError

client = ATOMClient(
    api_key="lcac_your_key_here",
    tenant_id="your-tenant-id"
)

# Make a governed call
result = client.governed_call(
    "Analyze this customer complaint...",
    provider="groq"
)

if client.is_allowed(result):
    print(client.get_output(result))
    print(f"RIS Level: {client.get_ris_level(result)}")
    print(f"CII Score: {client.get_cii_score(result)}")

# Raise exception if blocked
try:
    output = client.governed_call_or_raise(
        "Your prompt here",
        provider="anthropic"
    )
except GovernanceBlockError as e:
    print(f"Blocked: {e.reason}")
    print(f"RIS: {e.ris_level}, CII: {e.cii}")

Methods

Method	Description
governed_call()	Make a governed AI call, returns full envelope
governed_call_or_raise()	Call or raise GovernanceBlockError on block
is_allowed(result)	Check if decision is allow
is_blocked(result)	Check if decision is block
get_ris_level(result)	Get RIS-0 through RIS-4
get_cii_score(result)	Get CII float 0.0–1.0
get_output(result)	Get model response text
get_drift_score(result)	Get semantic drift score
get_traces()	Get recent governance traces
get_usage()	Get tenant usage stats
health()	Check platform health
create_agent()	Create a governed agent
list_agents()	List tenant agents
run_agent()	Execute a governed agent run
get_agent_policy()	Read agent governance policy
set_agent_policy()	Update agent governance policy

| get_webhooks() | List registered webhook endpoints | | create_webhook(url, events) | Register a new webhook endpoint | | delete_webhook(webhook_id) | Remove a webhook endpoint | | invite_user(email, role) | Invite a team member (admin/developer/viewer/security_officer) | | list_users() | List all team members for this tenant |

Team Management

# Invite a team member
client.invite_user("[e-mail]", role="developer")

# List current team
users = client.list_users()
for user in users:
    print(f"{user['email']} — {user['role']}")

Roles: admin · developer · viewer · security_officer

Webhooks

# Register a webhook endpoint
client.create_webhook(
    url="https://your-app.com/atom-webhooks",
    events=["governance.block", "governance.warn", "governance.allow"]
)

# List registered webhooks
hooks = client.get_webhooks()

# Delete a webhook
client.delete_webhook(hook_id)

Payloads are signed with HMAC-SHA256. Verify the X-ATOM-Signature header.

Local Models

# Use a local GGUF model (zero data egress)
result = client.governed_call(
    "Your prompt",
    provider="local",
    model="llama-3.2-3b-instruct.Q4_K_M.gguf"
)

Local models run on your infrastructure. No data leaves your environment.

Providers

groq · anthropic · gemini · mistral · openai · local

CLI

# Health check
python -m atomlabs health --api-key lcac_... --tenant-id your-tenant

# Make a governed call
python -m atomlabs call --input "Your prompt" --provider groq \
  --api-key lcac_... --tenant-id your-tenant

# Or via environment variables
export ATOM_API_KEY=lcac_...
export ATOM_TENANT_ID=your-tenant
python -m atomlabs health
python -m atomlabs call --input "Your prompt"
python -m atomlabs traces --limit 10
python -m atomlabs agents

Documentation

https://console.atomlabs.app/docs