Skip to main content
connic

connic vs OpenAI Agents APIAgents on any model, or OpenAI's Codex harness

The Agents API runs OpenAI's Codex harness for long tasks that your application starts and supplies with tools. connic runs agents on any model, starts them from inboxes, queues, databases and phone calls, and keeps platform data in an EU region.

Sources and methodologyLast reviewed

connic and the OpenAI Agents API at a glance
connicAgents API
Built for
Agents that react to business eventsLong tasks on the Codex harness
Agents and tools
YAML and code in your codebaseAPI config; function tools run in your app
Models
Managed, your own key or endpointModels from OpenAI's API
What starts a run
Email, queues, databases, calls and moreAPI calls and scheduled automations
Releasing a version
Test gate and live A/B testsSaved-agent updates for new sessions
Approvals
Per tool, with reviewer notificationsRequired actions or a reviewer model
Data residency
EU region you pickUnited States only

The short answer

The Agents API suits teams on OpenAI models that want a managed Codex harness for long tasks, with tools and approvals handled by their own application. connic suits teams whose agents should react to email, queues, databases and calls, run on the model they choose, keep data in the EU and pass tests before each release.

Use connic when

  • Inbound email, Kafka, SQS, database or payment events should start agents
  • Each agent should run on the model you pick, from any provider
  • Agent data has to stay in the EU
  • New versions should pass tests and prove themselves on live traffic first
  • Reviewers should approve tool calls in a dashboard, with notifications

Use the Agents API when

  • Your agents run on OpenAI models and you want OpenAI's own harness
  • Agents need a sandbox to run code, edit files and use a browser
  • Your application should run the tools and steer long sessions
  • A reviewer model should approve or deny supported requests on its own
  • Data residency in the United States meets your requirements

Where connic and the OpenAI Agents API differ

What each platform does in the areas that usually decide an evaluation.

Models

connic

Each agent names its model: a connic/* model running on EU capacity, a model from your own provider key, or your own endpoint.

Agents API

The documentation shows OpenAI models, and model usage is billed at OpenAI's API rates. Other providers are not documented.

How agents are built

connic

An agent is a YAML file with its model, prompt and tools. Custom tools are functions in your code that connic runs, and connic dev syncs every save to a dev environment.

Agents API

An agent has a model, instructions, tools and MCP servers, saved through the API or passed in when a session starts. Function tools run in your application, which returns each result to the session.

What starts a run

connic

Connectors for inbound email, Kafka, SQS, Postgres, storage and Stripe events, webhooks, cron, phone calls and messaging, defined in code or in the dashboard.

Agents API

Your application starts sessions and sends tasks, and scheduled automations run once or on a cron schedule. Webhooks report when the agent finishes or needs input.

Sandbox and tools

connic

Built-in tools search the web, read pages and use a browser, next to your own tools and MCP servers. connic documents no code interpreter.

Agents API

Sessions can use an OpenAI-hosted sandbox or one you host, where the agent runs code, edits files and connects to MCP servers.

Releasing a new version

connic

The agent's tests run before a deployment goes live, and it activates only if every case passes. A/B tests send a share of real traffic to a variant and pause when a safety guardrail is breached.

Agents API

Updates to a saved agent apply to new sessions. A test gate and traffic splits are not documented, and the Evals platform shuts down on November 30, 2026.

Approvals

connic

Approvals hold a tool call under conditions you set per tool, until someone decides in the dashboard, through the API or in your own app. Reviewers get email, in-app and webhook notifications.

Agents API

Browser access to a new site waits as a required action until your application decides. With automatic review, an OpenAI reviewer model approves or denies supported requests.

Quality and traces

connic

Span-by-span traces with token cost per run. Judges score a sample of finished production runs against a rubric.

Agents API

Traces of model responses, tool calls and subagents in the dashboard, exportable as OTLP JSON, with best-effort usage per session.

Hosting and EU

connic

Managed runtime with environments, and self-hosting on enterprise. Each project keeps its platform data in the EU region picked at creation, and the contract is German.

Agents API

Hosted service with OpenAI-hosted or self-hosted sandboxes. Data residency is available only in the United States, and Zero Data Retention is not supported.

The same agent in connic and the OpenAI Agents API

A support agent with one tool that looks up an order. In connic it is a YAML file and a Python function that connic runs. In the Agents API it is a session request with the tool's schema, plus code in your application that answers each required action.

connic

agents/order-support.yaml
name: order-support
model: connic/gpt-6-astra
system_prompt: Check the order first.
tools:
  - orders.get_order
tools/orders.py
import os
import httpx

API = os.environ["ORDER_API_URL"]

async def get_order(order_id: str) -> dict:
    """Look up an order by number.

    Args:
        order_id: Order number from the message
    """
    url = f"{API}/orders/{order_id}"
    async with httpx.AsyncClient() as client:
        response = await client.get(url)
        return response.json()

Agents API

app.py
import json
import os

import httpx
from openai import OpenAI

API = os.environ["ORDER_API_URL"]
client = OpenAI()

session = client.beta.agents.sessions.create(
    agent={
        "model": "gpt-6-astra",
        "instructions": "Check the order first.",
        "tools": [{
            "type": "function",
            "name": "get_order",
            "description": "Look up an order by number.",
            "parameters": {
                "type": "object",
                "properties": {"order_id": {"type": "string"}},
                "required": ["order_id"],
                "additionalProperties": False,
            },
        }],
    },
    environment={"type": "none"},
    input="Where is order 1042?",
)


# Your webhook calls this on agent.session.requires_action
def answer(session_id: str):
    sessions = client.beta.agents.sessions
    pending = sessions.retrieve(session_id)
    for action in pending.required_actions:
        order_id = action.arguments["order_id"]
        url = f"{API}/orders/{order_id}"
        order = httpx.get(url).json()
        sessions.events.create(session_id, events=[{
            "type": "agent.session.input.tool_result",
            "turn_id": action.turn_id,
            "call_id": action.call_id,
            "success": True,
            "output": json.dumps(order),
        }])

connic vs OpenAI Agents API, feature by feature

Each row is scored from connic's and OpenAI's public documentation. The methodology and sources follow the table.

  • YesFull support
  • PartialLimited, beta or extra setup
  • NoNot available
FeatureconnicAgents API
Build & runtime
Agent definitions in your codebaseconnic agents are YAML files with tools in code, deployed from your codebase. Agents API agents are configurations saved through the API or passed in when a session starts.YesPartial
Custom tools run on the platformconnic runs your tool functions as part of the deployment. In the Agents API, a function call reaches your application as a required action, and your code runs it and sends the result back. Attaching an environment does not run function tools there.YesNo
Sandbox with shell, files and codeAgents API sessions can run in an OpenAI-hosted or self-hosted sandbox where the agent executes code, edits files and connects to MCP servers, and computer use drives a browser. connic's built-in web tools search, read pages and use a browser, and connic documents no code interpreter.PartialYes
Models from other providersconnic agents use a connic/* model, a model from your own provider key, or your own endpoint. The Agents API documentation shows OpenAI models billed at OpenAI's API rates and names no other providers as of November 2026.YesNo
MCP servers as toolsBoth give agents tools from MCP servers.YesYes
Triggers & channels
Scheduled runsScheduled automations, added in October 2026, run an Agents API task once or on a five-field cron schedule. connic's cron connector starts runs on a schedule.YesYes
Email, queue and database triggersconnic connectors start runs from inbound email, Kafka and SQS queues, Postgres events, storage events and Stripe payment events. Agents API sessions start from your application or a scheduled automation, and other event triggers are not documented as of November 2026.YesNo
Phone callsconnic's voice connectors answer phone calls through providers such as Twilio and Telnyx, with the same agents and tools. OpenAI offers voice through its separate Realtime and Live APIs, with SIP calling, outside the Agents API.YesPartial
Release & quality
Tests gate each deploymentconnic runs the agent's tests before a deployment goes live and activates it only if every case passes, with rollback to an earlier deployment. Updates to a saved Agents API agent apply to new sessions, and a test gate is not documented.YesNo
Judges on production runsconnic judges score a sample of finished production runs against a rubric. OpenAI shuts down its Evals platform on November 30, 2026 and recommends migrating to Promptfoo, and scoring of production sessions is not documented for the Agents API.YesNo
Live A/B testsconnic sends a share of real traffic to a variant and pauses the test when a safety guardrail for quality, reliability, cost or latency is breached. Traffic splits between agent versions are not documented for the Agents API.YesNo
Run history and tracesThe Agents API records model responses, tool calls and subagent activity as traces in the dashboard, exports them as OTLP JSON and reports best-effort usage per session. connic traces every run span by span, with token cost per run.YesYes
Approvals
Human approval before a tool runsIn the Agents API, browser access to a new site waits as a required action until your application decides, and automatic review lets an OpenAI reviewer model approve or deny supported requests such as shell commands, file changes and MCP tool calls. connic holds any tool call with its parameters, under conditions you set per tool, until someone decides.YesPartial
Approvals in a dashboard, with notificationsconnic reviewers decide in the dashboard, through the API or in your own app, and get notified by email, in-app and webhook. The Agents API sends a webhook when a session needs input, and your application shows the approval controls.YesPartial
Hosting & EU
EU data residencyEach connic project keeps its platform data in the EU region picked at creation, and connic/* models run on EU capacity. The Agents API supports data residency only in the United States, although other OpenAI API endpoints offer an EU region.YesNo
Self-hosted deploymentAgents API sessions can use a sandbox you host, while OpenAI runs the service. connic offers self-hosting on enterprise.PartialPartial
Pricing
Published platform ratesThe Agents API bills model usage at the selected model's API rates, and OpenAI-hosted sandboxes at container rates, from $0.03 for a 1 GB container per 20-minute session. connic publishes base rates before package discounts: €0.0337 per started LLM-agent run, €0.00042 per runtime second and €0.00342 per GB-hour of storage.YesYes

Methodology

We compare generally available capabilities documented in each vendor's product, pricing, security, and legal material. Yes means documented support, Partial means limited scope, beta or preview status, or additional setup, and No means unavailable or not documented in the sources reviewed. connic capabilities are checked against current product documentation, and roadmap claims are not scored.

The Agents API is in public beta, so its rows describe the beta as documented in November 2026. Agent Builder and the Evals platform shut down on November 30, 2026 and are not scored. The Agents SDK, which runs in your own infrastructure, is mentioned where it fills a gap but is not scored as part of the hosted API.

Last reviewed November 15, 2026 by the connic research team.

Frequently Asked Questions

Yes, for agents that should react to events in your systems, run on the model you choose and keep their data in the EU. connic runs agents defined in YAML and code, starts them from inbound email, queues, databases and phone calls, and tests each deployment before it goes live. The Agents API fits long tasks on OpenAI models that need a sandbox and that your application starts and supplies with tools. The connectors page shows what can start a run.

No. The Agents API currently supports data residency only in the United States, and it is not eligible for Zero Data Retention. Other OpenAI API endpoints offer an EU region, at a 10% uplift for models released since March 5, 2026. connic keeps each project's platform data in the EU region picked at creation, and connic/* models run on EU capacity. Its catalog marks the managed GPT models as coming from a provider that is not EU-native, with 30-day provider retention. The EU data residency shortlist compares the other platforms on the same points.

OpenAI announced on June 3, 2026 that Agent Builder and the Evals dashboard, API and graders shut down on November 30, 2026. For Agent Builder it recommends the Agents SDK or ChatGPT Workspace Agents, and ChatKit stays available. For Evals it recommends migrating to Promptfoo. In connic, tests and judges are part of the platform: tests gate each deployment, and judges score sampled production runs.

Model usage is billed at the selected model's API rates, OpenAI tools at their standard rates, and OpenAI-hosted sandboxes at container rates, from $0.03 for a 1 GB container per 20-minute session. Automatic review adds its own tokens. connic developer and pro start at €40 and €200 per month, and the plan amount becomes Project credit at €0.0337 per started LLM-agent run, €0.00042 per runtime second and €0.00342 per GB-hour of storage, before package discounts. connic pricing lists every plan.

Not directly. connic runs agents defined in its own YAML format with tools in code, and it does not deploy agents built with other frameworks. Moving an agent means writing its prompt, model and tools as a connic agent, and its function tools become plain functions in your code.