--- title: "AI Agent Integration Not Working? Check These" url: https://www.insulin.dev/blog/ai-agent-integration-not-working/ canonical: https://www.insulin.dev/blog/ai-agent-integration-not-working/ type: Blog description: "AI agent integration not working in Insulin? Ask the agent first: it tells a missing integration from one that's connected but failing. Causes and fixes." --- # AI Agent Integration Not Working? Check These > Canonical HTML version: https://www.insulin.dev/blog/ai-agent-integration-not-working/ 1. [Home](/) 2. [Blog](/blog/) 3. AI Agent Integration Not Working? Check These # AI Agent Integration Not Working? Check These When an Insulin agent says it can't reach a system, ask it before you reconnect anything: it can tell an integration that was never added from one that's connected but failing. Each symptom, its cause, and where the fix is. ![Shirley Guo](/authors/shirley-guo.jpg) Shirley Guo Oct 23, 2026 ![AI Agent Integration Not Working? Check These](/images/blog/ai-agent-integration-not-working/hero.png) Explore AI Summary [![](/logos/company/openai.svg)](https://chat.openai.com/?q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fai-agent-integration-not-working%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Integrations%2C%20Agents. "Summarize with ChatGPT")[![](/logos/company/anthropic.svg) ](https://claude.ai/new?q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fai-agent-integration-not-working%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Integrations%2C%20Agents. "Summarize with Claude")[![](/logos/company/gemini.svg)](https://www.google.com/search?udm=50&aep=11&q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fai-agent-integration-not-working%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Integrations%2C%20Agents. "Summarize with Gemini")[](https://www.perplexity.ai/search/new?q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fai-agent-integration-not-working%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Integrations%2C%20Agents. "Summarize with Perplexity") Table of Contents - [Where should you look first when an agent’s integration isn’t working?](#where-should-you-look-first-when-an-agents-integration-isnt-working) - [Symptom, cause, fix: the diagnostic table](#symptom-cause-fix-the-diagnostic-table) - [Is the integration missing, or connected but failing?](#is-the-integration-missing-or-connected-but-failing) - [Why can’t an organization agent use your own connections?](#why-cant-an-organization-agent-use-your-own-connections) - [Why does a connected agent read but not change anything?](#why-does-a-connected-agent-read-but-not-change-anything) - [Why does an organization agent have no Fours tools?](#why-does-an-organization-agent-have-no-fours-tools) - [What’s different when you ask in Slack?](#whats-different-when-you-ask-in-slack) - [What if a scheduled job failed on an integration?](#what-if-a-scheduled-job-failed-on-an-integration) - [What does “Connection expired” mean?](#what-does-connection-expired-mean) - [Frequently asked questions](#frequently-asked-questions) - [Takeaways](#takeaways) _An AI agent integration that isn’t working is one of two problems: the agent can’t reach the connected system, or it reaches it but isn’t allowed to do what you asked. In Insulin, the agent can tell you which, because every agent can list its own integrations, check whether each one is working and list the actions it may call._ * * * “Why can’t you update Salesforce?” When an agent answers that it can’t reach a system, the instinct is to open **Settings → Integrations** and reconnect. Often that’s the wrong fix. The integration may never have been added to the agent, the agent’s scope may rule that connection out, or the connection may be fine and its access level may stop short of the change you asked for. Reconnecting fixes only one of those: a connection whose credential has stopped working. This post is for the person who owns an agent, or whose agent says it can’t reach something: three questions to ask the agent, a table of symptoms, then each cause in turn. ## Where should you look first when an agent’s integration isn’t working? **Ask the agent, before you open Settings or reconnect anything.** Every Insulin agent can inspect the connections it has been given, with no control to switch on, so the first diagnosis happens in the same chat. Because “is it attached?” and “does it still work?” are answered separately, the agent tells **not connected** apart from **connected but failing**, “instead of reporting a live-but-erroring connection as disconnected and sending you to reconnect something that is already connected,” as [Fours’ Agents documentation](https://doc.fours.com/insulin/agents/#asking-an-agent-about-its-own-integrations) puts it. Paste these three, in order, with the system that failed in place of Salesforce: 1. _List your integrations, and say whether each one is user-level or organization-level._ The answer says whether the system is attached to this agent at all, and at which level. 2. _Check whether Salesforce is working._ This runs a live probe, which refreshes an expired OAuth sign-in if it can, then reports the connection healthy or unhealthy, with the reason it failed. 3. _List the Salesforce actions you can call, and say which ones aren’t read-only._ The answer is what this agent may call at that integration’s access level, in the kind of turn it’s running, plus non-secret details such as an account id or base URL. Credentials are never returned. The first answer settles whether to add an integration, the second whether to reconnect one, and the third whether to change a level. ## Symptom, cause, fix: the diagnostic table **Find what the agent said, or what you saw, in the first column; the other two give the cause and where the fix is.** What you see What it means What to do The agent says the system isn’t connected, or leaves it out when it lists its integrations It isn’t on this agent’s allowlist, or nobody has connected it under **Settings → Integrations** Connect it if it isn’t connected, then add it to the agent’s **Integrations**. On an organization agent, only an org admin can An organization agent can’t use your Gmail or Outlook Organization agents use organization-level integrations only, and Outlook has no organization-level connection Ask a personal agent or the Insulin assistant, or have an org admin connect an organization Gmail with **Sign in with Google** and add it The Insulin assistant, or the Slack bot, says it doesn’t have an integration your organization connected The assistant uses your own connections; an organization-level connection doesn’t count Connect the system under your own account. For an organization-only system such as BigQuery, ask an organization agent that has it The agent reports a connection **unhealthy** Connected but failing: its credential stopped working, or, for example, a Salesforce sign-in is missing your instance address Reconnect it under **Settings → Integrations**. It stays on the agent’s list An organization integration’s card reads _Connected, but this integration’s stored credential can no longer be used_ Fours re-checked the credential against the provider and it failed, so agents and workflows can’t reach the system An org admin opens the card’s **⋯** menu, chooses **Details** and clicks **Reconnect** It reads but won’t create or change anything The integration is at **Read-only** for this agent Raise it to **Edit**, if writing there is the agent’s job It changes records but won’t delete or archive one The integration is at **Edit**, which never deletes **Full**, only if deleting is the job (below) The BigQuery or Databricks query tool is missing `run-query` and `run-sql` run any SQL statement, so an agent gets them only with **Full** **Full**, if querying is the job; the warehouse credential’s own permissions set what the SQL can change Your own Fours MCP connection won’t create or change anything Its **Access** is **Read only**, the default; a connection made before the choice existed counts as Read only too Disconnect it under **Custom MCP Servers** and connect it again with **Read & write** For one turn, the assistant wouldn’t write anywhere Fours couldn’t read your levels when the turn started, so it ran the turn with every integration at **Read-only** That’s the fail-safe, not your setting. Check the levels in the assistant’s panel, then ask again An organization agent has no Fours tools No org admin has added the Fours MCP server to it, or the person it’s working for isn’t an organization ADMIN or EDITOR Ask an org admin to add the server In a Slack channel nothing happened, and the reply ends _No approval within 3 minutes — nothing was sent._ The approval card waited three minutes for you, the only person who can answer it Ask again, then answer the card In Slack, a tool or an action you use in the web workspace is missing In a channel, Insulin doesn’t use an MCP server you connected yourself, and in Slack and Teams it acts without org-admin rights Ask in a DM for your own MCP server’s tools. Actions that need an org ADMIN aren’t available from Slack or Teams A scheduled job failed on an integration A job runs as the agent that created it, with that agent’s integrations and levels Open the failed run, which opens on **Steps**, and read the tool call’s **Error**. Fix the agent, not the job A Claude Code or Codex card shows _Connection expired — reconnect to keep using it_ That sign-in expired, or was revoked at the provider Click **Reconnect** on the card; there’s no need to disconnect first ## Is the integration missing, or connected but failing? **A missing integration was never added to the agent, or never connected; a failing one is attached, but its credential no longer works.** You add the first and reconnect the second, and the agent’s check is what tells them apart. **An unhealthy connection is** one the agent’s live probe could not use, reported with the reason it failed. There’s a third state between the two. “A connection can also be connected but unusable — for example, a Salesforce connection whose stored sign-in is missing the address of your Salesforce instance.” The agent reports that one as unhealthy too: it’s connected, but none of its tools are available in that run, and the fix is to reconnect it in **Settings → Integrations**. Organization-level integrations get a second check, on their cards. Each time you open **Settings → Integrations**, Fours re-checks each organization connection’s stored credential against the provider, and a card whose credential no longer works says so: “Connected, but this integration’s stored credential can no longer be used — agents and workflows cannot reach it.” The fix, per the [connection health notes in Fours’ Integrations documentation](https://doc.fours.com/integrations/#connection-health), is the card’s **⋯** menu, then **Details**, then **Reconnect**. Two limits apply to that card: - **A plain Connected isn’t proof.** The check runs on a short budget, and a card reads **Connected** whether it passed or couldn’t be checked in time. When an agent’s run fails anyway, ask the agent to check the connection. - **An API key is checked for presence, not validity.** For a connection made with an API key, the check confirms the key is still stored but doesn’t ask the provider whether it works, so a key revoked or rotated at the provider isn’t caught there. Reconnect with the new key. **Why do OAuth sign-ins stop working?** Because the provider decides how long a sign-in lasts. Microsoft’s identity platform is a documented example, and Outlook connects through Microsoft OAuth. Microsoft’s [page on refresh tokens](https://learn.microsoft.com/en-us/entra/identity-platform/refresh-tokens) says they “automatically expire once the lifetime period elapses” and “can be revoked by the sign-in service at any time before their expiration,” and in its revocation table, a user revoking their refresh tokens, or an admin revoking all of a user’s, ends every kind it lists. The agent’s check refreshes an expired sign-in when it can; once the provider has ended the grant, only a reconnect brings it back. ## Why can’t an organization agent use your own connections? **Because organization agents use organization-level integrations only, and personal connections don’t count.** Fours’ getting-started guide states the rule in two sentences: “Organization-level agents can only use organization-level integrations. User-level agents use user-level integrations.” A shared agent reaches the same systems whoever is chatting with it, which is the point; [how to restrict which integrations an AI agent can access](/blog/scope-an-agent-to-its-integrations/) explains why that boundary matters. Who’s asking Which connections it can use A personal agent Your user-level integrations selected for that agent An organization agent Organization-level integrations selected for that agent The built-in Insulin assistant, in the web workspace or in Slack Your own connections. An organization-level connection doesn’t count Three systems show the rule at work: - **Gmail** connects at both levels: an organization mailbox that everyone’s workflows share, or each person’s own. An organization agent gets only the organization’s. - **Outlook** is connected per user only, so no organization agent can use it. Ask a personal agent or the assistant. - **BigQuery** is an org-level integration only, so neither the assistant nor a personal agent can query it. Ask an organization agent that an org admin has given BigQuery. The same rule explains why two people can get different answers from the assistant, in Slack or anywhere else: each runs on their own connections. As the Slack bot documentation puts it, “That is the model working, not a fault.” Fixing an organization agent falls to an org admin. Only org admins create organization agents, and on one, “editing integrations requires organization administrator access,” so a teammate with EDITOR on the agent who isn’t an org admin can change its prompt and model but not its integrations. Before creating one, make sure the integrations it needs are connected at the organization level. ## Why does a connected agent read but not change anything? **Its access level for that integration stops short of the action.** **An access level is** the setting on each integration an agent uses, and there are three: “Read-only looks things up, Edit also changes records, Full also deletes them.” [Agent access levels](https://doc.fours.com/integrations/#agent-access-levels) in Fours’ Integrations documentation sets out what each gives. The agent won’t It’s at It needs Create or change a record **Read-only** **Edit** Delete, archive, trash, void or revoke something **Edit** **Full** Run a BigQuery or Databricks query **Read-only** or **Edit** **Full** **Full is the default,** and it’s what an integration selected before the setting existed keeps. So if an agent won’t write somewhere, look first for a level someone set below Full. On a custom agent, it’s the select beside each integration in the agent’s **Integrations** list. For the Insulin assistant, click the **pencil** in a chat’s header and use the panel’s **Integrations** section; those levels narrow Insulin for you, “not for anyone else in the organization.” **Edit never deletes,** by design. What counts as a delete, and when Full is worth granting, is the subject of [how to stop an AI agent from deleting records](/blog/stop-ai-agent-deleting-records/). **The query tools need Full.** BigQuery’s `run-query` “passes whatever SQL it is given to BigQuery, statements that change data included, so an agent gets it only with **Full** access.” With Read-only or Edit, the agent can still list datasets and tables and read their schemas. Databricks’ `run-sql` follows the same rule, and its other nine tools, which only read, stay available at every level. With Full, what the SQL can change inside the warehouse is up to the connection’s own credential: for BigQuery, the permissions granted to its service account. [Where read-only access for AI agents is set](/blog/read-only-access-for-ai-agents/) covers narrowing a credential to reading. To see which actions change data, open an integration’s **Details** from its card’s **⋯** menu in **Settings → Integrations**. The **Actions** tab marks each one **Read-only** or **Mutating**, and “a Mutating action needs **Edit**, or **Full** when it removes something.” The third prompt above gives the same answer for one agent, at its level. Two cases look like a level you set, but aren’t one: - **The fail-safe.** “If Fours cannot read your levels when a turn starts, Insulin runs that turn with every integration at **Read-only** rather than risk an action you restricted.” It applies to the assistant’s levels, for that turn. - **Your own Fours MCP connection.** On agents it carries a badge instead of a select, because its level belongs to the connection: **Read only**, **Read & write** or **Read, write & delete** on the connection card, shown on agents as **Read-only**, **Edit** or **Full**. **Read only** is the default, and a connection made before the **Access** choice existed counts as Read only, “so it gets no write tools until you reconnect it with wider access.” To change it, disconnect the server under **Settings → Integrations → Custom MCP Servers** and connect it again. The level only narrows: Fours still checks your own permissions on every call. ## Why does an organization agent have no Fours tools? **Either no org admin has added the Fours MCP server to it, or the person it’s working for isn’t an organization ADMIN or EDITOR.** An organization agent gets tools for your Fours records, offers, products, entitlements, co-sell referrals and more, through the Fours MCP server, which an org admin adds to the agent’s integrations; the list shows it under its former name, as `Suger MCP`. It’s offered only on organization agents and only to org admins. Even once it’s added, “its tools are attached only when the person the agent is working for is an organization **ADMIN** or **EDITOR**; for anyone else, the agent runs without them.” So the same agent can have Fours tools for one colleague and none for another. Its level is fixed. On an organization agent, the server carries a fixed **Edit** badge: the agent works on the organization’s own credential and can read and change records, “but never delete them.” A personal agent or the assistant reaches Fours through your own Fours MCP connection instead, at the level you chose when you connected it. In a Slack DM, a reply that mentions no Fours data connection means that per-person connection is missing. ## What’s different when you ask in Slack? **In a Slack channel, anything Insulin can’t recognize as a read waits for your approval, and Insulin doesn’t use an MCP server you connected yourself; in Slack and Teams alike, it acts without org-admin rights.** Each one can look like a broken integration. - **The approval card expired.** Creating or changing anything, and posting a Slack message, pause on a card that only the person who asked can answer. If nobody answers within three minutes, nothing is sent, and the reply ends _No approval within 3 minutes — nothing was sent. Ask again if you still want it._ In a DM, only posting a Slack message waits. - **Your own MCP server’s tools are missing.** In a channel, Insulin doesn’t use an MCP server you connected yourself. Ask in a DM instead. - **An admin action isn’t there.** In Slack and Teams, Insulin acts without organization-admin rights, even when you’re an org ADMIN, “so actions that need an ADMIN aren’t available there.” ## What if a scheduled job failed on an integration? **Fix the agent, not the job: a job runs as the agent that created it, with that agent’s access to integrations.** Ask the Insulin assistant for a job and it runs as the assistant, on your connections; ask a custom agent and it runs with that agent’s integrations and levels. A failed run opens on its **Steps** tab, the tool-call transcript, where each call shows its **Args** and its **Result** or **Error**, so the integration that failed and the error it returned are on the run. For a job that didn’t run at all, see [what to check when a scheduled AI job doesn’t run](/blog/why-a-scheduled-ai-job-did-not-run/). ## What does “Connection expired” mean? **It’s the alert on a personal Claude Code or Codex connection when Fours finds its sign-in no longer works, because it expired or you revoked it at the provider.** A green check on an integration card means only that a connection exists, so for those two sign-ins Fours “verifies the stored credential itself rather than trusting the card.” When the check fails, the card shows an orange alert, _Connection expired — reconnect to keep using it_, and a **Reconnect** button that runs the same connect flow you used before. You don’t need to disconnect first, and the alert clears once the new connection is in place. - **Only Claude Code and Codex show it.** They’re sign-ins to AI models, which a personal agent can run on, rather than systems an agent acts on. Connections made with an API key, such as Anthropic and OpenAI, “cannot be verified without using the key, so they never show this state.” - **No alert isn’t proof of health.** If Fours can’t reach the credential store, the check is skipped and the cards stay as they are, rather than telling everyone to reconnect. ## Frequently asked questions ### Why does my AI agent say an integration isn’t connected? Usually it isn’t on that agent’s list, or the agent can’t use that kind of connection: an organization agent uses organization-level integrations only, and the Insulin assistant uses only your own. Ask the agent to list its integrations. An attached but failing connection is reported as unhealthy instead. ### Should I reconnect an integration when my agent can’t use it? Only when the agent reports it unhealthy, or its card says the stored credential can no longer be used. If the integration isn’t on the agent’s list, or its access level blocks the action, reconnecting changes nothing. Ask the agent to check the connection first. ### Why can my agent read a system but not update it? Its access level for that integration is Read-only, which removes the write tools. Edit adds creating and changing records, and Full adds deleting. An integration nobody narrowed runs at Full. Your own Fours MCP connection is the exception: it defaults to Read only until you reconnect it. ### Why doesn’t my agent have the BigQuery query tool? BigQuery’s run-query tool passes any SQL it’s given to BigQuery, statements that change data included, so an agent gets it only with Full access. With Read-only or Edit, it can still list datasets and tables and read their schemas. Databricks’ run-sql follows the same rule. ### What does “Connection expired — reconnect to keep using it” mean? Your personal Claude Code or Codex sign-in no longer works: it expired, or you revoked it at the provider. Click Reconnect on the card; you don’t need to disconnect first. Connections made with an API key, such as Anthropic and OpenAI, never show this alert. ### Why did nothing happen when I asked Insulin in a Slack channel? In a channel, anything Insulin can’t recognize as a read waits on an approval card only you can answer. After three minutes with no answer, nothing is sent, and the reply ends “No approval within 3 minutes — nothing was sent.” Ask again, then answer the card. ## Takeaways - **Ask the agent before you reconnect.** It lists its integrations, checks whether each one works and lists the actions it may call, so it tells not connected from connected but failing. - **Reconnect only what’s failing.** An unhealthy connection, or a card whose stored credential can no longer be used, needs a reconnect; a missing integration needs adding. - **Scope decides which connections count.** Organization agents use organization-level integrations only, and the Insulin assistant uses only your own. - **A level isn’t a fault.** Read-only can’t write, Edit can’t delete, and the BigQuery and Databricks query tools need Full. Full is the default; your own Fours MCP connection defaults to Read only. - **Fours tools on an organization agent take two things:** an org admin adding the Fours MCP server, and an asker who is an organization ADMIN or EDITOR. - **An unanswered card in a Slack channel sends nothing** after three minutes, and the reply says so. An agent that can describe its own connections is the first place to look when one stops working. To see how Insulin agents are scoped to the systems they need, with a level on each integration, start with [custom AI agents in Insulin](/agents/); the [Agents documentation](https://doc.fours.com/insulin/agents/#integrations-and-knowledge-bases) is the reference for every control named here. ## Sources Primary sources for the platform rules cited above. Last verified October 23, 2026. Providers change their products, prices and terms without notice — check the source before relying on a detail. - [Insulin Agents — Fours Doc](https://doc.fours.com/insulin/agents/) — Asking an agent about its own integrations: every agent able to list its integrations (user- or organization-level, and when connected), get one integration's details, check whether one is working with a live probe that refreshes an expired OAuth sign-in if it can and reports it healthy or unhealthy with the reason, and list the provider actions it may call given the access level and the kind of turn, with whether each is read-only and no credentials returned; no control to switch on; not connected told apart from connected but failing; a Salesforce connection whose stored sign-in is missing the instance address reported unhealthy, with none of its tools available and reconnecting in Settings → Integrations the fix. Integrations and Knowledge Bases: user-level, org-level and system agents' integration access; Read-only, Edit and Full, Full the default and kept by integrations selected before the setting existed; the Fours MCP server added to an organization agent by an org admin, listed under its former name, Suger MCP, with a fixed Edit badge and its tools attached only when the person the agent works for is an organization ADMIN or EDITOR; your own Fours MCP connection's badge; editing an organization agent's integrations requiring org admin access; only org admins creating org-level agents. What the Insulin assistant may do with your integrations: the pencil panel's Integrations levels, yours alone, and a turn run with every integration at Read-only when Fours cannot read the levels. Slack and Teams: no organization-admin rights there; in a channel, actions it can't recognize as reads waiting for the asker's approval and no MCP server you connected yourself used; the three-minute expiry and the reply text No approval within 3 minutes — nothing was sent - [Insulin Getting Started — Fours Doc](https://doc.fours.com/insulin/getting-started/) — When a connection expires: Fours verifying the stored credential for personal Claude Code and Codex connections, the orange alert Connection expired — reconnect to keep using it, Reconnect running the same connect flow without disconnecting first, API-key connections such as Anthropic and OpenAI never showing the state, and the check skipped when the credential store can't be reached. Sharing constraints: organization-level agents can only use organization-level integrations, user-level agents user-level integrations - [Integrations — Fours Doc](https://doc.fours.com/integrations/) — Connection health: organization-level cards re-checked against the provider each time Settings → Integrations opens, the message Connected, but this integration's stored credential can no longer be used, reconnecting from the card's ⋯ menu → Details → Reconnect, a plain Connected card meaning passed or not checked in time, and API-key connections checked only for a stored key. Agent access levels: Read-only, Edit and Full, an integration never narrowed running at Full, the Actions tab marking each action Read-only or Mutating, a Mutating action needing Edit, or Full when it removes something - [BigQuery integration — Fours Doc](https://doc.fours.com/integrations/bigquery/) — BigQuery an org-level integration only, on a GCP service account with granted BigQuery permissions; run-query passing whatever SQL it is given, statements that change data included, so an agent gets it only with Full; Read-only or Edit keeping list-datasets, list-tables and get-table-schema - [Databricks integration — Fours Doc](https://doc.fours.com/integrations/databricks/) — run-sql running whatever statement it is given, so an agent gets it only with Full access; Read-only or Edit keeping the other nine tools, which only read - [MCP (Model Context Protocol) integration — Fours Doc](https://doc.fours.com/integrations/mcp/) — Connect Fours' own MCP server: the Access levels Read only (default), Read & write and Read, write & delete, shown on agents as Read-only, Edit and Full; a connection made before the Access choice existed counting as Read only; changing the level by disconnecting and connecting again; Access never widening what your Fours account may do - [Insulin Slack Bot — Fours Doc](https://doc.fours.com/insulin/slack-bot/) — Troubleshooting: It says it does not have an integration meaning the tool is not connected under your account, an organization-level connection not counting; It mentions no Fours data connection meaning the per-person Fours MCP connection is missing; two people asking the same question getting different answers because each runs on their own connections - [Insulin Jobs — Fours Doc](https://doc.fours.com/insulin/jobs/) — A job running as the agent that created it, with that agent's access to resources and integrations; a failed run opening on Steps, the tool-call transcript with each call's Args and its Result or Error, persisted when a run fails - [Gmail integration — Fours Doc](https://doc.fours.com/integrations/google-mail/) — Org-Level vs User-Level: an organization mailbox that everyone's workflows share, or each user's own Gmail connected via OAuth 2.0 - [Microsoft Outlook integration — Fours Doc](https://doc.fours.com/integrations/microsoft-outlook/) — Org-Level: not available, Outlook connected per user only; User-Level: each user authenticating via Microsoft OAuth - [Microsoft Learn: Refresh tokens in the Microsoft identity platform](https://learn.microsoft.com/en-us/entra/identity-platform/refresh-tokens) — Refresh tokens automatically expiring once the lifetime period elapses and revocable by the sign-in service at any time before their expiration; the revocation table, in which a user revoking their refresh tokens or an admin revoking all of a user's refresh tokens revokes every class of token. Page last updated November 5, 2025 ## Keep reading - [IntegrationsAI Agent for Jira in Slack: What Waits for YouOct 19, 2026](/blog/ai-agent-for-jira-in-slack/) - [IntegrationsAI Agent for Stripe: Payouts, Invoices, TaxOct 15, 2026](/blog/ai-agent-for-stripe/) - [IntegrationsAI Agent for BigQuery: Let It Query, Not ChangeOct 7, 2026](/blog/ai-agent-for-bigquery/) - [IntegrationsAI Model Retirement: What Happens to Your AgentsOct 7, 2026](/blog/ai-model-retirement/) [Browse every post on the Insulin Blog](/blog/) ### Stay Updated New posts, product updates and marketplace strategy are shared on LinkedIn as they publish. [Follow Fours on LinkedIn](https://www.linkedin.com/company/suger-inc)