Agent troubleshooting

Fixes for common agent problems: an agent that isn't posting, posts stuck in review, usage limits, failed or paused routines, disconnected accounts, and posts that are too long.

On this page20 sections

Why hasn't my agent posted?

Check these in order:

  1. Posts are waiting for your review. In Pilot mode nothing goes out until you approve it, and posts waiting for review don't hold a queue slot. Open the agent's chat, find the posts card and click Review posts or Add all to queue.
  2. The queue is empty. Open the agent's side panel (click its name at the top of its chat), Queue tab. It shows the next two weeks on that platform, your own posts and the agent's, with open slots as dashed boxes. If there are no posts, ask the agent for some (for example "Plan the week and fill my queue").
  3. The weekly routine is off or paused. On the Info tab, under Routines, check that the routine is active. If you chose Only when I ask at setup, the agent only writes when you message it.
  4. The account is disconnected. If the agent said its account needs reconnecting, or a post says Waiting for LinkedIn (or X, Threads, Substack, Instagram), reconnect the account (see "My agent says my account is disconnected").
  5. A post failed. Look for Did not publish on the Queue tab, click the post, then Requeue or Publish now.
  6. You're out of AI usage, or your plan changed. See "My agent says I've used my weekly AI allowance" and "Why did my agent's routine not run?".
  7. Your Digital Brain is thin. Agents only repurpose your own writing. If there isn't enough material, the agent leaves slots open rather than inventing posts. Add more to your Digital Brain (see the Digital Brain help).

Why are my agent's posts stuck in review?

The agent is in Pilot mode, so every post waits for you. Posts waiting for review:

  • Show "Needs your review" on their card.
  • Don't appear on the Queue tab and don't hold a slot.
  • Never publish on their own.

To clear them, open the agent's chat, find the posts card and click Review posts (or Add all to queue). Approved posts take your next open slots.

If you'd rather not approve each post, switch the agent to Autopilot: side panel, Info tab, Mode. In Autopilot, new posts are approved by default and publish on schedule.

Why didn't my agent write new posts when I still have posts in review?

When the agent plans, each post from the past week that is still waiting for your review counts as filling one slot. This stops the agent from writing a second batch on top of posts you haven't seen. Review the waiting posts, and the next plan fills whatever is still open.

My agent says I've used my weekly AI allowance

You see "You have used your weekly AI allowance. Open Settings → Usage to add credit or enable auto-reload." Agents use AI each time they reply, plan or design.

To keep going:

  1. Click your name at the bottom of the left sidebar and choose Settings (on a phone, tap your picture at the top).
  2. Open AI Usage.
  3. Buy extra usage, or turn on auto-reload.

Work that stopped because of the limit can be retried after you top up. If you see "Your monthly top-up spending limit has been reached", raise the limit in the same AI Usage settings. See the billing help for details.

My agent says "Agents are part of All-Access"

Agents are included in the All-Access plan, not Basic. If you're on Basic (or moved to Basic), agents can't reply or run their routines. Upgrade to All-Access from Settings, Billing.

"An active Typeshare plan is required. Choose a plan to continue." means you have no active plan. Choose one from Settings, Billing.

Why did my agent's routine not run?

If a scheduled routine fails, the routine shows "The last scheduled run failed" (on the agent's Info tab, under Routines) with the reason, and you get an email. Common reasons:

  • "you're out of AI usage for now": add usage in Settings, AI Usage (button Manage usage).
  • "you don't have an active Typeshare plan" or "you're on Basic, and agents are part of All-Access": fix your plan in Settings, Billing.
  • "the AI model was temporarily unavailable": the routine first tries again by itself 15 minutes later (twice at most), so you only get this if those tries failed too. After that it runs again at its next scheduled time.
  • "the agent ran into an error": it tries again at its next scheduled time.

A routine set to When I publish waits about 2 minutes after you publish, so every platform has the post first; the chat shows when it will run. It runs once per post, only for the formats it is set to, and only for posts published from Typeshare. A routine set to When I schedule runs about a minute after you schedule a post in its formats; if you unschedule the post first, it doesn't run.

After 3 failed runs in a row, the routine is paused and shows "Paused after 3 failed runs in a row". Fix the cause, then open the routine and click Turn back on.

My agent says my account is disconnected

When an agent's account (X, LinkedIn, Threads, Substack, Instagram or Pinterest) is disconnected, its session ends, or the platform stops accepting the login (you removed Typeshare's access on the platform, say), nothing posts until you reconnect. The agent tells you in its chat ("Disconnected LinkedIn: publishing is paused until it is reconnected", or that the login stopped working), and a line above the message box says the account isn't connected, with a button to reconnect:

  • The agent keeps working: you can chat with it, its routines still run, and Add to queue and Add all to queue still work.
  • Post now and Publish now are off until you reconnect.
  • When a queued post's time comes, it waits instead of failing: it shows Waiting for LinkedIn on the Queue tab and on your Schedule. The first time this happens after a disconnect, we email you once with a link to reconnect.

To reconnect:

  1. Open the agent's chat and use the line above the message box: click Use @yourhandle (or Reconnect), or Other account to use a different one. You can keep chatting with the agent meanwhile. Or go to Settings, Channels, and reconnect it there. A platform you had connected stays listed under Add a channel even when its login was removed.

For Substack, "Your Substack session has ended" means you need to reconnect through the Typeshare Connector extension in Chrome: with the extension added, log in to substack.com in that browser, then go to Settings > Channels and click Reconnect on Substack. If the Connector window says "Substack didn't accept the login in this browser", log out of Substack and back in, then click Try again. If it still fails, ask Ty to send a support request, or use Help & feedback in the account menu.

As soon as the account is connected again, every post that was Waiting for LinkedIn goes back in your queue, oldest first, in your next open slots. The chat says how many went back.

My Substack Agent said my session ended, but Settings > Channels shows Substack connected. Can it publish?

Yes, when Channels shows Substack connected. Opening Channels checks your login with Substack again: if Substack still accepts it, Substack is marked connected again, publishing carries on, and the agent's chat adds a "Connected substack" row under the earlier "session ended" one. You don't need to reconnect. Substack sometimes refuses the stats of one Note (for example a Note posted from another Substack account) while your login works; that only stops stats for that Note and no longer pauses your agent. If Channels says the session ended, reconnect as above.

Does my agent still work while my account is disconnected?

Yes. You can chat with it, its routines run and it fills your queue as usual. Only posting stops: posts wait until you reconnect.

What happens to my scheduled posts if my account disconnects?

They wait. A post whose time comes while the account is disconnected shows Waiting for LinkedIn (or X, Threads, Substack, Instagram) instead of failing, and you get one email about it. When you reconnect, the waiting posts go back in your queue in your next open slots, and posts still scheduled for later keep their times. You can also delete a waiting post, or edit it, from the Queue tab.

Why is my X post too long?

X posts are limited to 280 characters, or 4,000 characters if your connected X account has X Premium. If a post is too long, shorten it in the Edit window or ask the agent to split it into a thread.

LinkedIn posts can be up to 3,000 characters and Substack Notes up to 4,000.

The agent stopped replying or shows an error in the chat

  • "The AI model is very busy right now. Wait a minute, then send your message again.": too many people are using the AI model at once. Wait about a minute and send your message again. Nothing is wrong with your account or agent.
  • "The model is temporarily unavailable. Please try again.": wait a moment and send your message again.
  • "The AI model’s safety filter blocked this message, so the agent couldn’t reply.": the AI model refused what was sent, not because of your account. Sending it again unchanged gets the same answer. Rephrase your message, or leave out the text, link or file that may have set it off, then send it again. In a chat with one agent, your next message starts a fresh session, and when your text was what got blocked, the agent no longer sees it or anything before it. See "Why did the AI model's safety filter block my message?".
  • "Our AI provider is temporarily unavailable. We’ve been alerted. Please try again shortly.": the AI service Typeshare uses stopped answering for everyone for a while. It has nothing to do with your plan or your AI usage, and you are not charged for it. We get an alert and fix it; try again in a little while.
  • "Something went wrong on our side. Send your message again and I'll pick up where we left off.": send your message again.
  • "Reconnecting to this conversation…": the chat is catching up with the agent. It takes a moment; if it doesn't finish within a few seconds, the message box unlocks anyway so you can send a message.
  • "The connection to the agent dropped. Check your connection, then reconnect.": your browser lost its connection while the agent was replying (Wi-Fi changed, the laptop slept). Your message or answer still reached the agent if it had started replying. Once you're back online, click Reconnect to see the reply, or reload the page. If you answered a question card and it opened again, answer it again.
  • "This request expired. Send a message to pick up where you left off.": you answered an approval card after the conversation behind it had ended (for example, the page was open a long time). Send a message to carry on; the agent picks up with what came before. Answers to the agent's other questions go through on their own: your answer is sent as a message and the agent carries on.
  • A Manage usage link under the error: you're out of AI usage (see "My agent says I've used my weekly AI allowance").

To stop a reply that is taking too long, click the square Stop button.

What does the "input-token limit (40M) per session" card in an agent chat mean?

A very long chat can show a card saying "This session has hit the input-token limit (40M) per session. This is a guardrail against defective long-running sessions." with Approve and Stop. It is a safety check that pauses a chat after a very large amount of AI use, in case something went wrong (an agent stuck in a loop, say). It isn't an error with your account. A chat you use every day can reach it after a few weeks of normal use.

  • Approve ("Grant a fresh token budget"): the agent carries on in the same chat. Nothing is lost or reset: it keeps the whole conversation, its memory of you, your posts, drafts and settings. Only the safety counter starts over. If the chat looks normal to you, choose this.
  • Stop ("Stop now"): the reply that was about to run is cancelled. Your chat history, posts and anything already saved stay. Your next message then goes to a fresh session of the same chat.

Pick Approve or Stop on the card, or type A, B, "Approve" or "Stop". Or just send your message as usual: it goes to a fresh session of the same chat, and the agent answers it there. The card has no Skip. AI used after you approve counts toward your AI usage as usual.

My agent says "Your queue has no open slot in the next 90 days"

Every slot in your posting schedule is taken. Either click Publish now on the post, or add more days or times to the agent's schedule: side panel, Info tab, Queue, then pick a preset with more posts or Custom… to set your own days and times.

The agent can't find my Notion page

Share the page with the Typeshare integration in Notion, then ask the agent again. If Notion isn't connected yet, connect it from Marketplace, Plugins, Notion, Connect.

The agent didn't add an image to my post

Agents add media to about half their posts by default (see Media on posts on the agent's Info tab; if it's set to None, they only add media when you ask). They use only your Media library or a shared Google Drive folder, and only a picture that fits the post. Agents never have the Designer make a picture for a post on their own. To get one, open the post and choose Make a visual from the image button's menu, or select some text and click Make a visual. A Designer visual made elsewhere has to be saved to Media first (Save to Media) before an agent can use it, unless you are in a channel with the Designer.

If your library has nothing that fits, the agent's icon appears next to the image button on the post. Hover it to see the picture the agent wanted. Add one like it to Media, then pick Suggest media from the image button's menu.

To get an image on a post, ask directly, for example "Add the whiteboard photo from my Photos folder to Wednesday's post", or attach the image in the agent's chat and ask the agent to use it.

Why did the AI model's safety filter block my message?

The AI model behind your agents has its own safety filter. Now and then it refuses a message, or a file or page the agent read, that it judges to be off limits, and the chat shows "The AI model’s safety filter blocked this message, so the agent couldn’t reply." Typeshare doesn't choose what it blocks, and it has nothing to do with your plan or AI usage.

  1. Rephrase your message in plainer words.
  2. If you shared a file, link or pasted text, leave it out or send only the part the agent needs.
  3. Send it again.

In a chat with one agent, your next message after a block starts a fresh session with the agent. When the filter blocked the text you sent, the agent no longer sees that message or the conversation before it, so later messages aren't blocked because of it. The earlier messages stay on screen, but the agent doesn't remember them, so repeat anything it still needs, such as which post you were working on. When it blocked an image or the agent's own reply instead, the agent still remembers the earlier conversation. Your posts, drafts, settings and memory are not affected.

If ordinary messages keep getting blocked, report it (see "How do I report a problem?" in the getting help docs).

Will a long agent chat ever stop and ask me to approve more AI use?

Rarely. Behind the scenes, each agent chat runs in a session, and a session pauses with an "input-token limit (40M) per session" card after a very large amount of AI use, as a safety check against an agent stuck in a loop. A chat you use every day used to reach it after a few weeks.

Now, before a chat gets close to that limit, your next message quietly starts a fresh session for the same chat. Nothing changes on your screen: same chat, same messages, no card. The agent picks up your recent conversation and its memory of you, and your posts, drafts and settings stay as they are. In a very long chat, it may not recall details from far back in the conversation, so remind it of anything older that it still needs.

If the card does show (one reply that used an unusually large amount of AI at once), you don't have to answer it: just send your message as usual, and it goes to a fresh session of the same chat. Approve and Stop on the card still work too. If the card keeps coming back, report it (see "How do I report a problem?" in the getting help docs).

Why did an agent chat take a few seconds to open?

Behind the scenes, each agent chat runs in a session. A chat used heavily for days (long replies, lots of back and forth) could take 10 seconds or more to open, because the app read that whole session every time.

Now the chat opens with only the latest part of the session, and your next message in a chat that long quietly starts a fresh session for the same chat, so it opens fast from then on. Nothing changes on your screen: same chat, same messages. The agent picks up your recent conversation and its memory of you, and your posts, drafts and settings stay as they are. It may not recall details from far back in the conversation, so remind it of anything older that it still needs.

If a chat still takes a long time to open, report it (see "How do I report a problem?" in the getting help docs).

What happens to my message if it doesn't send?

If a message to an agent fails to send, the chat shows "That message did not send. Try again." and puts your message back in the message box, with any images or files you attached, so you can send it again without retyping it. If you've already started typing something new by then, your new text stays and the failed message isn't put back. If messages keep failing, see "The agent stopped replying or shows an error in the chat" above.

Still stuck? Ask Ty, bottom right. Or email help@typeshare.co.

All help articles

Try it for yourself.

Free for 14 days

Try Typeshare free