OpenAI · Technologies
Agents SDK
Multi agent systems without the framework tax.
The Agents SDK is OpenAI’s open source framework for building agent systems in Python and TypeScript, with a deliberately small set of primitives: agents, handoffs and guardrails, plus built in tracing. It is the durable, code first path for agents on OpenAI, and it plays well beyond OpenAI too. We build production agent systems on it.
AGENT FLEET · EXAMPLE SYSTEM
AGENT
ROLE
HANDOFF
STATUS
Research-Agent
Gather
To writer
LIVE
Writer-Agent
Draft
To review
LIVE
Review-Agent
Check
To human
LIVE
Tracing
Observability
Enabled
REVIEW
rogue-experiment
Unknown owner
None
RISK
Sample system
live · review · risk
In plain terms
One agent is a demo. A governed fleet is a system.
The gap between an agent that impresses and one that ships is engineering. Here is what changes.
Without it
- Orchestration logic invented from scratch per project
- Agents passing work through fragile glue code
- No trace of why an agent did what it did
- Experiments multiplying with no owners
With it
- A small, proven set of primitives doing the orchestration
- Handoffs between specialist agents, by design
- Every run traced, debuggable and evaluable
- A registered fleet with owners and guardrails
What CG TECH can do with the Agents SDK
The work, broken into the parts that matter.
Power without the learning curve
Agents, handoffs and guardrails cover most real systems, in plain Python or TypeScript your team can read, test and version like any other code.
power without a steep learning curve
No handcuffs in the orchestration
The SDK works with OpenAI models by default and with other providers through compatible APIs, and MCP servers connect your systems once for reuse everywhere.
orchestration without vendor handcuffs
Agents that talk and act
Realtime voice agents with interruption handling, and computer use for the systems that only speak user interface, all inside the same framework.
agents that talk and act
Agents you can debug and prove
Built in tracing shows every step, evaluations measure improvement, and with OpenAI winding down its visual Agent Builder from November 2026, code first with the SDK is the durable path.
agents you can debug and prove
How an engagement runs
From a demo agent to a governed fleet, step by step.
01
Discover
We map the workflow, the systems involved and the risk profile.
02
Design
Agent roles, handoffs and guardrails designed before code.
03
Build
The system shipped with tracing and evaluation from day one.
04
Handover
Code, traces and runbooks your engineers own outright.
Questions we hear a lot
Common questions about the Agents SDK
What is the OpenAI Agents SDK?
The Agents SDK is OpenAI’s open source toolkit for building agent systems in Python and TypeScript. It keeps to a small set of building blocks, agents, handoffs and guardrails, with tracing built in, which makes it the code first option when you want agents you can version, test and maintain like any other software.
Agents SDK or Copilot Studio?
Different audiences. Copilot Studio is low code inside the Microsoft estate; the Agents SDK is pro code for custom systems. We build with both and recommend per workload.
We built workflows in OpenAI’s Agent Builder. What now?
OpenAI is winding it down, with availability ending from 30 November 2026, and recommends the Agents SDK for workflows that should live as code. We plan and run that migration.
Does it lock us into OpenAI?
No. The SDK is open source and works with other model providers through compatible endpoints, so the orchestration you invest in is portable.
How do we keep autonomous agents safe?
Guardrails on inputs and outputs, human approval on consequential actions, tracing on everything, and in a Microsoft estate, governance through Agent 365. Autonomy is earned in stages, not granted on day one.
Who is this for, developers or business teams?
Developers. If nobody in your organisation writes code, Copilot Studio or a packaged product will serve you better. The SDK is worth it when the logic is genuinely complex or the agent has to live inside a product you build.
What is a handoff?
It is one agent passing work to another that is better suited to it, the way a triage officer passes a case to a specialist. It matters because trying to make one agent do everything is the most common reason agent projects become unreliable.
How do we test an agent properly?
The same way you test anything else: known inputs, expected outputs, and a record of what changed between versions. Tracing is built into the SDK, which is what makes that possible. Agents that cannot be tested should not be in production.
Ready when you are
Building a real agent system? Let us talk.
A discovery session maps your workflow, your risks and your quick wins. You keep the plan either way.
What to expect
- A consultant replies within 4 business hours
- Session booked to understand your requirements
- We will provide you with a fixed price quote