What is Dev Mode?

Dev Mode is a development mode for integration within ChatGPT, designed for testing features, APIs, and user scenarios before production release.

In this mode, the application may operate unstably — this is normal.

---

Why does the application sometimes not “see” data?

ChatGPT does not maintain a persistent connection to our service.

Sometimes:

• authorization expires;

• the connection to the API is reset;

• the MCP/API connector is restarted;

• ChatGPT loses the current context/tool session.

Usually helps:

• reconnect the integration;

• refresh the chat;

• repeat the request.

---

Why does ChatGPT say that tools are unavailable?

In Dev Mode, tools may:

• temporarily disable;

• not load after deployment;

• fail to initialize in time;

• timeout.

This especially concerns:

• analytics API;

• streaming endpoints;

• heavy queries;

• long-running requests.

---

Why does the same request sometimes work and sometimes not?

Possible reasons:

• different tool sessions;

• recreation of the sandbox;

• rate limiting;

• API timeout;

• unstable Dev backend;

• context overflow.

In Dev Mode, stability is lower than in production.

---

Why does ChatGPT “forget” data?

Each request does not always execute in the same tool/session environment.

Because of this:

• state may be lost;

• temporary data may disappear;

• a repeated request may yield a different result.

---

Why doesn’t ChatGPT always call the integration?

ChatGPT independently decides:

• whether to call the tool;

• which tool to use;

• whether the connector is available;

• whether there is enough context.

Sometimes the model may:

• respond without the API;

• use old context;

• not call the tool in case of routing error.

---

Why do large requests perform poorly?

Dev Mode is not well-suited for:

• huge reports;

• heavy analytics;

• large JSON;

• long polling;

• massive datasets.

Better to:

• break down requests;

• use pagination;

• limit the period;

• request top-N data.

---

Why does ChatGPT sometimes provide incorrect data?

If the API is unavailable or returned an error:

• the model may attempt to respond based on context;

• use old data;

• interpret an incomplete response.

For critical data, it is always recommended to:

• verify the numbers;

• refresh the request;

• use short and specific commands.

---

How to properly use the integration?

It is recommended to:

• ask specific requests;

• specify the period;

• specify the team/GEO/campaign;

• avoid overly broad requests.

Good examples:

• “Show top GEO for today”

• “Compare installs yesterday and today”

• “Analyze revenue drop over 7 days”

• “Which campaigns dropped in CR?”

Bad examples:

• “What’s happening?”

• “Analyze everything”

• “Why is it flowing poorly?”

---

What to do if the integration stopped working?

1. Refresh the chat.

2. Reconnect the integration.

3. Restart Dev Mode.

4. Check the API separately.

5. Repeat the request later.

6. Create a new chat.