Name	Name	Last commit message	Last commit date
parent directory ..
ai	ai
general	general
README.md	README.md

OnceOnly Python SDK — Examples

Quick, runnable examples showing real-world OnceOnly patterns for
Execution Safety and AI Agent Governance (budgets, permissions, kill switch, audit).

🎯 Start here: quickstart.py — 3-minute copy-paste demo

🚀 Quick Start (60 seconds)

# 1. Install SDK
pip install -e .

# 2. Set API key
export ONCEONLY_API_KEY="once_live_..."

# 3. Run first example
python examples/quickstart.py

Expected output:

✅ First call: locked=True, duplicate=False
❌ Second call: locked=False, duplicate=True
✨ Idempotency working!

📁 Example Categories

Pick your use case:

Category	When to use	Start with
Webhooks & Workers	Stripe webhooks, cron jobs, background tasks	`webhook_dedup.py`
AI Agents	Long-running AI jobs, tool calling, function execution	`ai_simple.py`
Governance	Budget limits, permissions, kill switch, audit	`governance.py`
Decorators	Function-level exactly-once	`decorator_basic.py`
LangChain	Agent tool wrapping	`langchain_integration.py`

🔹 Webhooks & Workers

Core Pattern

Prevent duplicate webhook/event processing:

# examples/general/webhook_dedup.py
result = client.check_lock(
    key=f"stripe:webhook:{event_id}",
    ttl=7200  # 2 hours
)

if result.duplicate:
    return {"status": "already_processed"}

# Process event (runs exactly once)
handle_payment(event_id)

Run it:

python examples/general/webhook_dedup.py

All Examples

File	Concept	Run time
`basic_check_lock.py`	Minimal idempotency	1s
`webhook_dedup.py`	Stripe webhook pattern	1s
`with_ttl.py`	TTL behavior	1s
`with_metadata.py`	Debug metadata	1s
`fail_open_demo.py`	Graceful degradation	2s
`usage_stats.py`	Account info	1s

🤖 AI Agents

Core Pattern

Long-running AI job with automatic deduplication:

# examples/ai/ai_simple.py
result = client.ai.run_and_wait(
    key="report:daily:2024-01-27",
    ttl=1800,
    timeout=120.0
)

if result.status == "completed":
    print(result.result)

Run it:

python examples/ai/ai_simple.py

Tool Runner (Governed)

Use this when you want budgets/permissions enforced per tool call:

# examples/ai/tool_permissions.py (uses policies + tool execution)
res = client.ai.run_tool(
    agent_id="support-bot",
    tool="send_email",
    args={"to": "user@example.com", "subject": "Hi", "body": "Welcome"},
    spend_usd=0.02
)

if res.allowed:
    print("Executed:", res.result)
else:
    print("Blocked:", res.policy_reason)

Async version:

res = await client.ai.run_tool_async(
    agent_id="support-bot",
    tool="send_email",
    args={"to": "user@example.com", "subject": "Hi", "body": "Welcome"},
    spend_usd=0.02
)

All Examples

File	Concept	Run time
`ai_simple.py`	Basic AI job	2s
`run_and_wait.py`	Poll until completion	5-10s
`agent_action_local.py`	Exactly-once local execution	3s
`poll_status.py`	Manual polling	5s
`get_result.py`	Fetch completed result	1s
`agent_full_flow_no_onceonly.py`	Full LLM flow (no OnceOnly)	5s
`agent_full_flow_onceonly.py`	Full LLM flow (with OnceOnly)	5s

🛡️ Governance

⭐ Quick Start: Complete Demo

Governance includes four safety layers:

Layer	Purpose
Budget Limits	Prevent runaway AI spending
Tool Permissions	Control which tools an agent can call
Kill Switch	Instantly disable a misbehaving agent
Action Audit Log	Full forensic trail of agent activity

API Endpoints Map (Public)

Core: GET /v1/me, GET /v1/usage, GET /v1/usage/all, GET /v1/events, GET /v1/metrics
Idempotency: POST /v1/check-lock
AI Jobs: POST /v1/ai/run, GET /v1/ai/status, GET /v1/ai/result
AI Lease (local side-effects): POST /v1/ai/lease, POST /v1/ai/extend, POST /v1/ai/complete, POST /v1/ai/fail, POST /v1/ai/cancel
Governance (policies): POST /v1/policies/{agent_id}, POST /v1/policies/{agent_id}/from-template, GET /v1/policies, GET /v1/policies/{agent_id}
Governance (agents): POST /v1/agents/{agent_id}/disable, POST /v1/agents/{agent_id}/enable, GET /v1/agents/{agent_id}/logs, GET /v1/agents/{agent_id}/metrics
Tools Registry: POST /v1/tools, GET /v1/tools, GET /v1/tools/{tool}, POST /v1/tools/{tool}/toggle, DELETE /v1/tools/{tool}

Plan Differences (Pro vs Agency)

Pro: Governance is limited (no allowed_tools whitelist, no kill switch).
Agency: Full governance (tool whitelist + kill switch).

Policy Templates (Server Defaults)

Available templates:

strict
moderate
permissive
read_only
support_bot

Plan Limits (Defaults)

Plan	`check_lock` (make)	`ai` (runs)	Default TTL	Max TTL	Tools Registry Limit
Free	1K / month	3K / month	60s	1h	Not available
Starter	20K / month	100K / month	1h	24h	Not available
Pro	200K / month	1M / month	6h	7d	10 tools
Agency	2M / month	10M / month	24h	30d	500 tools

Tools Registry (Pro / Agency)

# Register a tool
tool = client.gov.create_tool({
    "name": "send_email",
    "url": "https://example.com/tools/send_email",
    "scope_id": "global",
    "auth": {"type": "hmac_sha256", "secret": "your_shared_secret"},
    "timeout_ms": 15000,
    "max_retries": 2,
    "enabled": True,
})

The best place to start — shows all governance features:

# examples/ai/governance.py

# 1. Set policy (budgets + permissions)
client.gov.upsert_policy({
    "agent_id": "billing-agent",
    "max_actions_per_hour": 200,
    "max_spend_usd_per_day": 50.0,
    "allowed_tools": ["stripe.charge", "send_email"],
})

# 2. Check metrics
metrics = client.gov.agent_metrics("billing-agent")

# 3. Kill switch
client.gov.disable_agent("billing-agent")
client.gov.enable_agent("billing-agent")

# 4. Audit logs
logs = client.gov.agent_logs("billing-agent")

Run it:

export ONCEONLY_AGENT_ID="billing-agent"  # optional
python examples/ai/governance.py

Expected output:

Setting policy...
Metrics: AgentMetrics(
    total_actions=15,
    blocked_actions=0,
    total_spend_usd=7.5,
    ...
)
Disabling agent...
Re-enabling agent...
Logs count: 15
First 3: [AgentLogItem(...), ...]

🔐 Tool Permissions

Whitelist/blacklist specific tools your agent can call:

# examples/ai/tool_permissions.py

client.gov.upsert_policy({
    "agent_id": "support-bot",
    "allowed_tools": ["send_email", "stripe.refund"],
    "blocked_tools": ["stripe.charge", "delete_user"],
})

Run it:

python examples/ai/tool_permissions.py

Expected output:

Setting tool permission policy...
Policy applied.

This agent can:
  ✓ send_email
  ✓ stripe.refund

This agent CANNOT call:
  ✗ stripe.charge
  ✗ delete_user

If the agent tries to call a blocked tool via `ai.run_tool`, you'll get `allowed=False` with a `policy_reason`.

When to use:

✅ Limit support agents to read-only + refund tools
✅ Prevent billing agents from deleting users
✅ Sandbox development agents

💰 Budget Limits

Set strict spending caps to prevent runaway costs:

# examples/ai/budget_limits.py

client.gov.upsert_policy({
    "agent_id": "support-bot",
    "max_actions_per_hour": 5,
    "max_spend_usd_per_day": 1.0,
    "allowed_tools": ["send_email"],
    "max_calls_per_tool": {
        "send_email": 2  # Only 2 emails per day
    }
})

Run it:

python examples/ai/budget_limits.py

Expected output:

Setting strict budget policy...
Policy set.
Metrics: AgentMetrics(total_actions=0, blocked_actions=0, total_spend_usd=0.0, ...)
Attempting to exceed limits (simulate)...
When limits are exceeded, API will return OverLimitError or 402.

When to use:

✅ Development/staging agents (tight budgets)
✅ Experimental agents (cap damage)
✅ Cost-sensitive production (prevent billing spikes)

What gets blocked:

# Agent tries 3rd email (max_calls_per_tool=2)
→ 402 OverLimitError: "Tool call limit exceeded"

# Agent exceeds $1/day
→ 402 OverLimitError: "Daily spend limit exceeded"

# Agent exceeds 5 actions/hour
→ 402 OverLimitError: "Hourly action limit exceeded"

All Governance Examples

File	Concept	Run time	Complexity
🌟 `governance.py`	Complete demo (start here)	3s	⭐ Beginner
`tool_permissions.py`	Whitelist/blacklist tools	2s	⭐ Beginner
`budget_limits.py`	Spending caps	2s	⭐ Beginner
`policy_templates.py`	Use pre-built templates	2s	⭐⭐ Intermediate
`kill_switch.py`	Emergency disable	2s	⭐⭐ Intermediate
`audit_logs.py`	Forensic investigation	2s	⭐⭐ Intermediate
`metrics_monitoring.py`	Usage metrics	2s	⭐⭐⭐ Advanced

📊 Governance Use Cases

Scenario 1: Support Bot (Strict Limits)

# examples/ai/budget_limits.py pattern
client.gov.upsert_policy({
    "agent_id": "support-bot",
    "max_actions_per_hour": 5,
    "max_spend_usd_per_day": 1.0,
    "allowed_tools": ["send_email"],
    "max_calls_per_tool": {"send_email": 2}
})

Scenario 2: Production Billing Agent

# High budget for production
client.gov.upsert_policy({
    "agent_id": "billing-prod",
    "max_actions_per_hour": 500,
    "max_spend_usd_per_day": 100.0,
    "allowed_tools": ["stripe.charge", "stripe.refund"],
    "blocked_tools": ["delete_user"]
})

Scenario 3: Development Agent (Sandbox)

# Strict limits for testing
client.gov.upsert_policy({
    "agent_id": "dev-agent",
    "max_actions_per_hour": 10,
    "max_spend_usd_per_day": 0.5,
    "blocked_tools": ["stripe.charge", "delete_user", "send_email"]
})

Scenario 4: Support Bot (Permission-based)

# examples/ai/tool_permissions.py pattern
client.gov.upsert_policy({
    "agent_id": "support-bot",
    "max_actions_per_hour": 100,
    "allowed_tools": [
        "get_user",
        "send_email", 
        "stripe.refund",
        "create_ticket"
    ],
    "max_calls_per_tool": {
        "stripe.refund": 5,   # Max 5 refunds/day
        "send_email": 50      # Max 50 emails/day
    }
})

Scenario 5: Emergency Stop

# Rogue agent detected
client.gov.disable_agent(
    "suspicious-agent",
    reason="Unusual spending pattern detected"
)

# Investigate
logs = client.gov.agent_logs("suspicious-agent", limit=100)
for log in logs:
    if log.spend_usd > 10:
        print(f"⚠️ High cost: {log.tool} - ${log.spend_usd}")

# Re-enable after fix
client.gov.enable_agent("suspicious-agent")

🎯 Decorators

Core Pattern

Function-level exactly-once execution:

# examples/general/decorator_basic.py

@idempotent_ai(
    client,
    key_fn=lambda user_id: f"welcome:{user_id}",
    ttl=86400
)
def send_welcome_email(user_id: str):
    email_service.send(...)
    return {"sent": True}

# Runs once, returns cached result on duplicates
send_welcome_email("user_123")  # Sends
send_welcome_email("user_123")  # Cached

Run it:

python examples/general/decorator_basic.py

All Examples

File	Concept	Run time
`decorator_basic.py`	Simple decorator	2s
`decorator_sync.py`	Sync functions	2s
`decorator_async.py`	Async functions	2s
`decorator_pydantic.py`	Pydantic models	3s

🔗 LangChain

Core Pattern

Wrap LangChain tools with idempotency:

# examples/ai/langchain_integration.py

from langchain_core.tools import tool
from onceonly.integrations.langchain import make_idempotent_tool

@tool
def send_email(to: str, subject: str) -> str:
    """Send email"""
    email_service.send(...)
    return f"Sent to {to}"

# Wrap with idempotency
safe_email = make_idempotent_tool(
    send_email,
    client=client,
    ttl=3600
)

# Use in agent
agent = create_react_agent(llm, tools=[safe_email], prompt)

Run it:

pip install langchain-core
python examples/ai/langchain_integration.py

All Examples

File	Concept	Run time
`langchain_integration.py`	Tool wrapping	3s
`langchain_tool_ai_lease.py`	Advanced patterns	5s

🎓 Learning Path

Never used OnceOnly? Follow this order:

1. quickstart.py              (understand basic concept)
   ↓
2. webhook_dedup.py           (real-world use case)
   ↓
3. ai_simple.py               (long-running jobs)
   ↓
4. governance.py              (⭐ complete governance demo)
   ↓
5. tool_permissions.py        (whitelist/blacklist tools)
   ↓
6. budget_limits.py           (spending caps)
   ↓
7. decorator_basic.py         (function-level safety)
   ↓
8. langchain_integration.py   (framework integration)

Already familiar? Jump to:

Full governance: governance.py ⭐
Tool permissions: tool_permissions.py
Budget enforcement: budget_limits.py
Kill switch: kill_switch.py
Audit logs: audit_logs.py

📊 Example Output Reference

Successful Execution

# First call
CheckLockResult(
    locked=True,
    duplicate=False,
    key="order:12345",
    ttl=3600,
    first_seen_at=None
)

# Duplicate call
CheckLockResult(
    locked=False,
    duplicate=True,
    key="order:12345",
    ttl=3600,
    first_seen_at="2024-01-27T12:00:00Z"
)

AI Job Completion

AiRun(
    ok=True,
    status="completed",
    key="report:daily:2024-01-27",
    result={"data": "..."},
    charged=1,
    usage=1,
    limit=1000
)

Policy Blocked

AiRun(
    ok=False,
    status="failed",
    error_code="policy_blocked",
    key="dangerous:action"
)

Budget Exceeded

# Over daily spend limit
OverLimitError: Daily spend limit exceeded ($1.00)
# HTTP 402

# Over tool call limit
OverLimitError: Tool call limit exceeded (send_email: 2/day)
# HTTP 402

# Over hourly actions
OverLimitError: Hourly action limit exceeded (5/hour)
# HTTP 402

Agent Metrics

AgentMetrics(
    agent_id="billing-agent",
    period="day",
    total_actions=127,
    blocked_actions=3,
    total_spend_usd=63.50,
    top_tools=[
        {"tool": "stripe.charge", "count": 95},
        {"tool": "send_email", "count": 32}
    ]
)

Agent Logs

# Allowed action
AgentLogItem(
    ts=1706356800,
    tool="stripe.charge",
    allowed=True,
    decision="executed",
    policy_reason="within_budget",
    risk_level="medium",
    spend_usd=0.5,
    args_hash="abc123..."
)

# Blocked action
AgentLogItem(
    ts=1706356850,
    tool="delete_user",
    allowed=False,
    decision="blocked",
    policy_reason="tool_not_allowed",
    risk_level="high",
    spend_usd=0.0
)

# Budget exceeded
AgentLogItem(
    ts=1706356900,
    tool="send_email",
    decision="blocked",
    reason="daily_spend_exceeded",
    risk_level="low",
    spend_usd=0.0
)

🔧 Customizing Examples

All examples use environment variables for easy customization:

# Required
export ONCEONLY_API_KEY="once_live_..."

# Optional
export ONCEONLY_BASE_URL="https://api.onceonly.tech/v1"
export ONCEONLY_TTL="3600"
export ONCEONLY_AGENT_ID="my-agent"

Per-Example Overrides

# Run governance example with different agent
ONCEONLY_AGENT_ID="support-bot" python examples/ai/governance.py

# Run with custom TTL
ONCEONLY_TTL="7200" python examples/general/webhook_dedup.py

# Run budget limits with different agent
ONCEONLY_AGENT_ID="dev-agent" python examples/ai/budget_limits.py

🐛 Troubleshooting

"No module named 'onceonly'"

Cause: SDK not installed

# From repository root
pip install -e .

"UnauthorizedError: Invalid API Key"

Cause: Missing or invalid API key

# Check key format
echo $ONCEONLY_API_KEY
# Should start with: once_live_

# Get key from dashboard
open https://onceonly.tech/dashboard/keys

"OverLimitError: Daily spend limit exceeded"

Cause: Agent hit budget cap (this is intentional!)

Debug:

# Check current usage
python examples/ai/governance.py

# View metrics
python -c "
from onceonly import OnceOnly
import os
client = OnceOnly(api_key=os.environ['ONCEONLY_API_KEY'])
metrics = client.gov.agent_metrics('support-bot', period='day')
print(f'Spend: \${metrics.total_spend_usd} / \$1.00')
print(f'Actions: {metrics.total_actions} / 5')
"

Solution: Increase budget or wait for daily reset:

client.gov.upsert_policy({
    "agent_id": "support-bot",
    "max_spend_usd_per_day": 5.0,  # Increase limit
})

"Rate limit exceeded"

Cause: Too many requests

Solution: Enable auto-retry:

client = OnceOnly(
    api_key=os.environ["ONCEONLY_API_KEY"],
    max_retries_429=3  # Auto-retry on 429
)

"ImportError: langchain_core"

Cause: LangChain not installed (optional dependency)

pip install langchain-core

"Policy not enforced"

Cause: Policy may not be applied yet or agent_id mismatch

Debug:

# Check if policy exists
try:
    metrics = client.gov.agent_metrics("my-agent")
    print("Policy active:", metrics)
except Exception as e:
    print("No policy found:", e)

# Check logs for blocked actions
logs = client.gov.agent_logs("my-agent", limit=10)
for log in logs:
    print(f"{log.tool}: {log.decision} - {log.reason}")

💡 Design Tips

Good Idempotency Keys

# ✅ Good: Deterministic, specific
key = f"stripe:webhook:{event_id}"
key = f"order:{order_id}:payment"
key = f"user:{user_id}:welcome_email"
key = f"report:daily:{date}"
key = f"agent:{agent_id}:action:{action_id}"

# ❌ Bad: Random, non-specific
key = f"webhook:{uuid.uuid4()}"       # Always unique
key = f"payment:{time.time()}"        # Never deduplicates
key = "generic_key"                   # Too broad

TTL Selection

# Webhooks: 2-24 hours (depends on retry window)
ttl = 7200  # 2 hours for Stripe

# Daily jobs: 24-48 hours
ttl = 86400  # 1 day

# Real-time events: 1-5 minutes
ttl = 300  # 5 minutes

# One-time migrations: Days/weeks
ttl = 604800  # 1 week

# AI agent actions: Match expected completion time
ttl = 3600  # 1 hour for typical tool calls

Metadata Best Practices

# ✅ Useful for debugging
meta = {
    "user_id": 123,
    "agent_id": "billing-bot",
    "source": "api",
    "trace_id": "abc-123",
    "amount_cents": 9999,
    "tool": "stripe.charge"
}

# ❌ Too much data
meta = {
    "full_request_body": {...},  # Too large
    "api_key": "...",            # Sensitive
    "password": "...",           # Never!
}

Budget Policy Design

# ✅ Good: Layered defense
policy = {
    "agent_id": "billing-agent",
    
    # 1. Rate limits (prevent runaway loops)
    "max_actions_per_hour": 200,
    
    # 2. Cost caps (prevent billing spikes)
    "max_spend_usd_per_day": 100.0,
    
    # 3. Tool permissions (allow only safe tools)
    "allowed_tools": ["stripe.charge", "send_receipt"],
    "blocked_tools": ["delete_user", "refund_all"],
    
    # 4. Per-tool limits (fine-grained control)
    "max_calls_per_tool": {
        "stripe.charge": 50,  # Max 50 charges/day
        "send_receipt": 100   # Max 100 emails/day
    }
}

# ❌ Bad: Too permissive
policy = {
    "agent_id": "billing-agent",
    # No limits = no protection!
}

# ❌ Bad: Too restrictive (blocks legitimate usage)
policy = {
    "agent_id": "support-bot",
    "max_actions_per_hour": 1,  # Too low
    "max_spend_usd_per_day": 0.01  # Unusable
}

Budget Sizing Guidelines

# Development/Testing
max_actions_per_hour = 10
max_spend_usd_per_day = 0.50

# Low-volume Production (support bot)
max_actions_per_hour = 50
max_spend_usd_per_day = 5.00

# Medium-volume Production (billing agent)
max_actions_per_hour = 200
max_spend_usd_per_day = 50.00

# High-volume Production (automation agent)
max_actions_per_hour = 1000
max_spend_usd_per_day = 500.00

📚 Additional Resources

Main README: ../README.md
API Docs: https://docs.onceonly.tech/reference/idempotency/
SDK Reference: https://docs.onceonly.tech/sdk/python
Dashboard: https://onceonly.tech/dashboard
Governance Guide: https://docs.onceonly.tech/governance
Budget Planning: https://docs.onceonly.tech/governance/budgets

🚦 Status Reference

HTTP	Error	Meaning	Action
200	—	Success	✅ Proceed
401	`UnauthorizedError`	Invalid API key	Check `ONCEONLY_API_KEY`
402	`OverLimitError`	Usage/budget limit	Upgrade plan or increase budget
403	`ApiError`	Forbidden / feature gating	Check plan + `error=feature_not_available`
422	`ValidationError`	Bad request	Fix parameters
429	`RateLimitError`	Rate limit	Enable retries
500+	`ApiError`	Server error	Retry or fail-open

🎯 Quick Reference

Basic Idempotency

result = client.check_lock(key="...", ttl=3600)
if result.duplicate:
    return "already_done"

AI Job

result = client.ai.run_and_wait(key="...", timeout=60.0)

Decorator

@idempotent_ai(client, key_fn=lambda x: f"task:{x}")
def my_task(x): ...

Governance (Budget Limits)

# Set strict budget
client.gov.upsert_policy({
    "agent_id": "my-agent",
    "max_actions_per_hour": 5,
    "max_spend_usd_per_day": 1.0,
    "allowed_tools": ["send_email"],
    "max_calls_per_tool": {"send_email": 2}
})

# Monitor usage
metrics = client.gov.agent_metrics("my-agent")
print(f"Spend: ${metrics.total_spend_usd}")

Governance (Audit Logs)

logs = client.gov.agent_logs("billing-agent", limit=20)
for log in logs:
    print(log.tool, log.decision, log.reason)

### Governance (Tool Permissions)
```python
# Whitelist safe tools
client.gov.upsert_policy({
    "agent_id": "support-bot",
    "allowed_tools": ["send_email", "stripe.refund"],
    "blocked_tools": ["delete_user"]
})

Governance (Kill Switch)

# Emergency disable
client.gov.disable_agent("rogue-agent", reason="suspicious")

# Audit
logs = client.gov.agent_logs("rogue-agent")

🌟 Featured Examples

Best examples to understand OnceOnly:

🥇 governance.py — Complete governance demo
🥈 budget_limits.py — Spending caps
🥉 tool_permissions.py — Whitelist/blacklist tools
🎖️ webhook_dedup.py — Classic idempotency

Start with governance.py to understand OnceOnly's full power.

Questions? Open an issue or email support@onceonly.tech

Found a bug? Submit a PR with a failing example!

_{Built with ❤️ by the OnceOnly team}

FilesExpand file tree

examples

Directory actions

More options

Directory actions

More options

Latest commit

History

examples

Folders and files

parent directory

README.md

OnceOnly Python SDK — Examples

🚀 Quick Start (60 seconds)

📁 Example Categories

🔹 Webhooks & Workers

Core Pattern

All Examples

🤖 AI Agents

Core Pattern

Tool Runner (Governed)

All Examples

🛡️ Governance

⭐ Quick Start: Complete Demo

API Endpoints Map (Public)

Plan Differences (Pro vs Agency)

Policy Templates (Server Defaults)

Plan Limits (Defaults)

Tools Registry (Pro / Agency)

🔐 Tool Permissions

💰 Budget Limits

All Governance Examples

📊 Governance Use Cases

🎯 Decorators

Core Pattern

All Examples

🔗 LangChain

Core Pattern

All Examples

🎓 Learning Path

📊 Example Output Reference

Successful Execution

AI Job Completion

Policy Blocked

Budget Exceeded

Agent Metrics

Agent Logs

🔧 Customizing Examples

Per-Example Overrides

🐛 Troubleshooting

"No module named 'onceonly'"

"UnauthorizedError: Invalid API Key"

"OverLimitError: Daily spend limit exceeded"

"Rate limit exceeded"

"ImportError: langchain_core"

"Policy not enforced"

💡 Design Tips

Good Idempotency Keys

TTL Selection

Metadata Best Practices

Budget Policy Design

Budget Sizing Guidelines

📚 Additional Resources

🚦 Status Reference

🎯 Quick Reference

Basic Idempotency

AI Job

Decorator

Governance (Budget Limits)

Governance (Audit Logs)

Governance (Kill Switch)

🌟 Featured Examples