Build and run AI agents on a European, code-first platform.
Engineering teams keep the agent definition, Python tools, and tests in Git. Connic deploys that repository to a managed EU runtime, with scaling, connectors, traces, storage, approvals, and governance built in.
Build your first agent- Build imagedone
- Run testsdone1/1 · 100%1/1
order-support::looks_up_an_order - Deploy to productiondone
tests/order-support.yaml runs before the production environment changes.What is a code-first AI agent platform?
Keep agent definitions, tools, and tests in Git. Connic deploys and operates them.
| Feature | Framework + assembled stack | Connic |
|---|---|---|
| Agent source | Framework-specific Python | YAML + ordinary Python |
| Multi-agent orchestration | Configure it in your application | Sequential and delegated agents |
| Deployment | Your CI/CD and runtime | Git- or CLI-triggered deployment |
| Tests before release | Your release gate | Tests run in the deploy pipeline |
| Production operations | Your operations stack | Managed execution and observability |
| EU operating boundary | Depends on selected services | EU Project region; EU inference for connic/* models |
Ship the agent from its repository
The mapped branch contains the agent YAML, Python tools, and release tests. Connic builds that commit and deploys it only after the tests pass.
version: "1.0"
name: order-support
type: llm
model: connic/glm-5.2
description: "Resolve order questions and delegate refunds"
system_prompt: |
Help customers understand an order.
Look up the order before answering.
Delegate refund requests to the refund agent.
tools:
- orders.lookup
- trigger_agent
guardrails:
input:
- type: prompt_injection
mode: blockimport os
import httpx
async def lookup(order_id: str) -> dict:
"""Return the current status of one order."""
async with httpx.AsyncClient() as client:
response = await client.get(
f"{os.environ['ORDERS_API_URL']}/orders/{order_id}",
headers={"Authorization": f"Bearer {os.environ['ORDERS_API_KEY']}"},
)
response.raise_for_status()
return response.json()version: "1.0"
tests:
- name: looks_up_an_order
payload: '{"message":"Where is order A-1042?"}'
expected_result: status == "completed"
expected_tool_calls:
- orders.lookup: invocations >= 1main → productionA push starts a deployment-gated release
Connic builds the commit, runs tests/order-support.yaml, and deploys the image after the suite passes. Non-Git Projects can deploy from the CLI.
After deployment, each execution is available as a run and trace with tool calls, latency, token use, and cost.
See how deployment-gated tests run, follow the deployment workflow, or build the complete Python example without Kubernetes.
Where Connic keeps processing in Europe, and where configuration matters
The EU boundary covers the selected Project region and connic/* managed inference. Customer-configured providers and destinations can send data beyond it.
- German company and contract
- Connic is based in Munich and contracts through a German entity. Procurement and legal teams can review the DPA and check the current subprocessor list.
- EU platform and managed inference
- Customers select an EU data region when creating a Project. Every
connic/*model request runs only on EU inference capacity and fails closed if that boundary cannot be maintained. - Customer-configured services
- End-to-end EU residency requires every configured model provider, tool, guardrail, judge, and external destination to stay in the EU. Check which models use managed EU inference.
- Governance records and evidence
- Enterprise AI Governance records controls and incidents and exports evidence for readiness work. Legal advice, final classification, and compliance certification remain outside the product. Review what AI Governance covers.
Read the product details
Each link opens the relevant product page or documentation.
Composer SDK
See how YAML agents and Python tools fit together.
Tools & orchestration
Add Python tools, built-ins, or MCP servers.
Models
Choose managed EU models or bring a provider.
Testing
Run assertions and judges before deployment.
Observability
Inspect runs, traces, latency, token use, and cost.
EU AI Act readiness
Record controls and incidents, then export evidence.