The control plane for AI agents
See every trace, enforce every policy, prove every compliance. One SDK. One dashboard.
import custos
@custos.agent(name="research-agent")
async def research(query: str) -> str:
prompt = custos.get_prompt("research-system")
return await llm.generate(prompt, query)Your agents send data. The Console gives you control.
You built your agents. Now who's governing them?
Blind execution
Your agents run, but you can't see what they're doing, which tools they called, or why they failed. No visibility means no accountability.
Compliance blind spots
The EU AI Act demands risk assessments, audit trails, and human oversight for high-risk AI systems. Without policy enforcement, every agent output is a potential regulatory violation.
Cost surprises
A runaway agent loop burns through your API budget before you notice. Without budgets and alerts, costs spiral unchecked.
Prompt chaos
System prompts live in code, with no versioning, no rollback, no audit trail. Regulators require traceable AI decision-making — scattered prompts make that impossible.
Designed for the EU AI Act
Every capability in Custos maps to specific regulatory requirements. From risk management to audit trails, your agents are built for compliance from day one.
Guardrails with block, warn, and log enforcement modes
Full execution tracing with searchable history
Trace Explorer with input/output inspection per step
Real-time dashboard with violation alerts and controls
Prompt Registry with version control and audit trail
Compliance Overview
On TrackPolicies active
0
Compliance score
0%
EU AI Act articles
Last audit: 2 min ago
Your mission control for AI agents
A self-hosted dashboard to monitor, govern, and optimize every agent in production. Deploy with docker compose up.
Dashboard
Real-time overview of your entire AI fleet. See active agents, trace volume, cost trends, and policy violations at a glance.
- Active agent count and status
- Daily trace volume with trend
- Cost trend with month-to-date total
- Policy violations with regulatory context
Active agents
0
Traces/day
0
Monthly cost
$0
Violations
0
Cost trend (30d)
Recent violations
How Custos works
Four pillars of agent governance — powered by the SDK, managed in the Console
Tracing
Full execution tracing with cost attribution per agent, per step. See every LLM call, tool invocation, and decision your agents make. Explore traces visually in the Console's Trace Explorer.
- Automatic span creation for every LLM call
- Cost attribution per agent and per step
- Latency breakdown across the execution chain
- Searchable trace history with filters
@custos.agent(name="research-agent")
async def research(query: str) -> str:
# Every call is traced automatically
result = await llm.generate(query)
return resultIntegrate in minutes
Add observability and guardrails to your existing agents with a few lines of code
import custos
custos.init(project="my-project", api_key="cs_...")
@custos.agent(name="research-agent", version="1.0")
async def research_agent(query: str) -> str:
# Fetch versioned prompt from registry
prompt = custos.get_prompt("research-system")
# LLM call — automatically traced with cost tracking
result = await llm.generate(
system=prompt,
user=query,
model="gpt-4o"
)
# Guardrails evaluate the output automatically
# Policies block, warn, or log based on your rules
return result@custos.agent()
Automatic tracing of every execution
get_prompt()
Versioned prompts with instant rollback
Guardrails
Policy enforcement with zero latency
Works with what you already use
Framework-agnostic by design. One SDK for any Python-based agent system.
Don't see your framework? Custos works with any Python-based agent system.
How Custos compares
The only platform combining tracing, active guardrails, prompt lifecycle, cost budgets, and regulatory compliance support.
| Feature | Custos | LangSmith | Langfuse | Arize Phoenix |
|---|---|---|---|---|
| Full execution tracing | ||||
| Active guardrails (block/warn/log) | ||||
| Prompt lifecycle (promote/rollback) | Basic | |||
| Cost budgets with alerts | ||||
| Governance dashboard | ||||
| Framework-agnostic | ||||
| Open-source SDK | ||||
| Self-hosted option | ||||
| Regulatory compliance support |
Full execution tracing
Active guardrails (block/warn/log)
Prompt lifecycle (promote/rollback)
Cost budgets with alerts
Governance dashboard
Framework-agnostic
Open-source SDK
Self-hosted option
Regulatory compliance support
Last verified: 2026-03
Simple, transparent pricing
Start with the right plan. Scale as your agents grow.
Free
For individuals and small teams getting started
- Console dashboard (self-hosted)
- 5,000 traces/month
- 2 projects
- 7-day data retention
- Community support
- Basic guardrails
- Prompt registry (10 prompts)
Pro
For teams running agents in production
- Full Console with all views
- 100,000 traces/month
- Unlimited projects
- 90-day data retention
- Priority support
- Advanced guardrails
- Policy editor
- Compliance policy templates
- Unlimited prompts
- Cost alerts & budgets
- Webhook notifications
Enterprise
For organizations with advanced requirements
- Managed cloud option
- Unlimited traces
- Unlimited projects
- Custom data retention
- Dedicated support
- Custom dashboards
- SSO / SAML
- RBAC & audit logs
- Compliance reporting & audit exports
- Custom regulatory framework support
- SLA guarantee
- On-premise option
Tell us what matters most in your agent stack
Share your priorities, the capabilities you care about, and the level of support you need. We will route the request with context instead of a cold generic inbox.
What happens next
- 01
We review your stack, use case, and required capabilities.
- 02
We reply with the right follow-up instead of a generic trial link.
- 03
We keep the request protected with CAPTCHA and server-side validation.
Good fit signals
- You need real trace visibility across agent workflows.
- Guardrails, budgets, or prompt lifecycle are already pain points.
- You want a self-hosted control plane instead of another black box.