OpenAI · Technologies
Responses API
The API OpenAI builds its own agents on.
The Responses API is OpenAI’s unified interface for agents and assistants, with tools like web search, file search, code interpreter and computer use built in, and remote MCP servers supported. It replaces the Assistants API, which shuts down on 26 August 2026. We build new work here and migrate the old.
AGENT BACKENDS · EXAMPLE ESTATE
INTEGRATION
API
STATE
STATUS
Support-Agent
Responses API
Current
LIVE
Web-Research
Built in tools
Current
LIVE
Order-Bot
Responses API
Current
LIVE
File-QA
Assistants API
Retiring
REVIEW
Doc-Assistant
Assistants API
Retiring
MIGRATE
Sample estate
live · review · migrate
In plain terms
One API call, tools included.
What used to take a stack of services now ships in a single call. Here is what changes.
Without it
- A different API for every capability, stitched by hand
- Web results, file retrieval and code execution all separate builds
- Agent integrations still sitting on a retiring API
- Little visibility into what the model did and why
With it
- One stateful API with the tools built in
- Web search, file search and computer use in a single call
- Everything off the Assistants API before the deadline
- Traces and evaluations showing exactly what happened
What CG TECH can do with the Responses API
The work, broken into the parts that matter.
Capability without the plumbing
Web search with citations, file search over your documents, code interpreter for analysis, computer use for interfaces and image generation, all callable in one request instead of five integrations.
capability without the plumbing
Your systems, one protocol away
The Responses API supports remote Model Context Protocol servers, and OpenAI sits on the MCP steering committee, so connections you build reach across vendors instead of locking to one.
your systems, one protocol away
Off the retiring API in good order
The Assistants API shuts down on 26 August 2026. We inventory what you built on it, map threads and data to their Responses equivalents and migrate with room to test.
migrated before the deadline, not after it
Agents you can inspect
Stateful conversations, tracing and evaluation support mean you can see what an agent did, measure whether it is improving and prove it to whoever asks.
agents you can inspect and improve
How an engagement runs
From a retiring API to a modern agent backend, step by step.
01
Audit
We inventory existing integrations, including anything on the Assistants API.
02
Design
Tools, state and guardrails designed for each workload.
03
Migrate and build
Old integrations moved, new capability shipped and tested.
04
Handover
Traces, evaluations and runbooks your team owns.
Questions we hear a lot
Common questions about the Responses API
What is the OpenAI Responses API?
The Responses API is OpenAI’s unified interface for building agents and assistants. Tools like web search, file search, code interpreter and computer use are built in, and remote MCP servers are supported. It replaces the Assistants API.
What happened to the Assistants API?
OpenAI replaced it with the Responses API and set a shutdown date of 26 August 2026. Anything still on it needs to move, and that migration is work we scope and run.
How is this different from a normal chat completion?
It is stateful and it carries tools. One call can search the web, read your files, run code and keep conversation state, instead of your team wiring each capability by hand.
Can it reach our internal systems?
Yes, through function calling and remote MCP servers. Build the connection once and it works across MCP aware tools, not just this one.
Is data we store with OpenAI used for training?
No. Business data is not used for training by default, including data stored to support these features.
When does the Assistants API shut down?
26 August 2026. If you have anything running on it, that is a hard deadline rather than a suggestion, and migration is the priority ahead of any new build.
How hard is the migration?
It varies with how much of the Assistants behaviour you leaned on. Straightforward assistants move quickly. The ones that take longer are those relying on threads and state in ways the new model handles differently. We would rather scope that now than in the last fortnight.
What is MCP and why does it matter here?
MCP is a common way for AI tools to connect to systems, so an integration you build can be reused rather than rewritten for each platform. Support for remote MCP servers means less bespoke plumbing between your agent and your systems.
Ready when you are
Building agents on OpenAI? Let us talk.
A discovery session maps your integrations, 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