Troubleshooting

Troubleshooting

This page lists the error messages and notices of AnyLLM verbatim — in the chat, in suggestion cards, in the history, in the settings, in the Confluence quick actions and in the dashboard gadget — with their cause and the solution. If you are looking for a message, use your browser's search on this page.

Check these first

  1. Can you see the chat at all? If not, or if it only shows a notice → The app does not start or is locked.
  2. Is the input field locked (placeholder Configuration by an admin required.)? Then the app is not set up yet → same section. (An inactive subscription does not lock the field; it only blocks sending — see below.)
  3. Does the error occur in Jira and in Confluence? Both products share one configuration, so a problem in only one product almost always means that the AI endpoint is not yet approved for that product (message "The endpoint … is not approved for … yet — endpoint approvals apply per product. …") or that Confluence is not connected to the app.
  4. Did it happen only once? Network-related interruptions occur. Send once more before reporting it.

The app does not start or is locked

MessageCauseSolution
Not set up yet with "The chat is available once an admin has configured the app."; input field shows Configuration by an admin required.no saved configuration, or no connection with a stored API keyadministration: Setting Up LLM Connections
"Not configured yet — an administrator sets the endpoint, models and API key in the app settings (Jira: Settings → Apps · Confluence: Settings → AnyLLM)." (chat banner)same as abovesame as above
"Not configured yet — an admin must set up connections and API keys in the app settings." (quick actions, chat bar answers, dashboard gadget)same as abovesame as above
"The AnyLLM subscription for this site is inactive — an administrator can review it in Atlassian administration (Billing)." (chat) · "… — an administrator can review it in Atlassian administration." (quick actions) · "… — an admin can review it in Atlassian administration." (server, e.g. in the gadget)the paid subscription is inactive. Only new answers and ✓ Execute are blocked; history, memory, own agents and skills and all settings keep workingadministration: check the subscription in Atlassian administration — see Licence, Data Retention and Automatic Deletion
"This feature has been disabled by your administrator." (in the page assistant, the issue action and the issue navigator with a Close link; in the gadget in view and edit mode; on the standalone full page)this particular surface is switched off under Modules; menu entries cannot be hidden per installationuse a different entry point, e.g. Apps → AnyLLM Chat; administration: Modules
⤢ Fullscreen on a Jira issue opens only the disabled noticethe module Fullscreen chat (Jira) is switched offadministration: Modules
"Page access for the AI is disabled by your admin — page actions are unavailable." (page quick actions greyed out)AI access is set to Offactions on selected text keep working; administration: General Settings
"Configuration could not be loaded."the settings were briefly unavailablereload the page; if it recurs, inform the administration
The chat bar in Confluence has disappearedDismiss was clicked (applies to the current view only), the surface is switched off, the app is not set up, the subscription is inactive, or the status could not be loaded — the bar shows no message in any of these casesreload the page. The bar's status is cached in your browser for ten minutes, so after the administration has switched it back on or completed the setup it can take up to ten minutes before the bar reappears
The Settings button is missing in the chatyou are not an administrator — in Confluence the button appears only for members of the usual administrator groupsopen the settings via the product: Jira Settings → Apps → AnyLLM Chat — Settings, Confluence Settings → AnyLLM Chat — Settings — see Opening the Settings and Initial Setup
The chat before configuration: the notice 'Not configured yet', the empty state 'Not set up yet' and the locked input field.

Messages while chatting

The request does not get through

MessageCauseSolution
"No response from stream (timeout) — please try again."nothing arrived from the stream for 60 secondssend again; if it recurs, inform the administration
"Stream error." / "Call failed (invocation error or timeout)."processing was interruptedsend again
"Hourly limit reached (… requests/hour) — please try again later."your personal hourly limit is reached. One counter covers Jira and Confluence together; every send counts (including ↻ Regenerate answer, quick actions and gadget refreshes); the window starts with your first request and ends one hour laterwait. The administration can adjust the limit — General Settings
"Please enter a prompt."an empty message was sententer some text
"Connection not configured or has been removed."the connection behind the chosen model was removed or has no API key any morechoose another model in the Model selector or reload; otherwise administration
"The endpoint … is not approved for … yet — endpoint approvals apply per product. An administrator can grant it in the AnyLLM settings of … under “Endpoint approvals”." (shown in your interface language)the AI address is not approved in this product — typical right after importing a configuration, or when it was only approved in the other productadministration: Managing Endpoint Approvals

The AI service refuses

MessageCauseSolution
"LLM endpoint rejects the API key (401/403)."API key wrong, expired or without permissionadministration: renew the key — Setting Up LLM Connections
"Rate limit of the LLM endpoint reached (429)."quota at the AI provider exhaustedwait a moment; if it persists, administration
"LLM endpoint responded with HTTP …" (followed by the service's reply)the service reports an erroradministration checks address, model names and quota
"Timeout after 120s — shorten the prompt or choose another model."one round of the model took longer than 120 secondsask a shorter question or choose another model
"LLM call failed: …", possibly with "This endpoint needs a network approval: check that an administrator has approved it under Settings → Endpoint approvals."the service could not be reached; with the addition, the address is not approvedadministration: Managing Endpoint Approvals
"The LLM response was empty or had an unexpected format."the service answered in a form the app does not understandadministration checks the provider type and the model names
"Internal stream error."an unexpected error while streamingsend again; if it recurs, inform the administration

The answer stays empty or breaks off

MessageCauseSolution
"The model used its token budget (8000) on internal reasoning without answering — shorten the prompt or choose a non-reasoning model."the model spent the whole response budget on internal reasoningask a shorter question or choose a model without an internal reasoning phase
"Model is thinking …" for a long timethe model is reasoning internally before text is producedwait; if this happens permanently, the administration may choose a different default model
"I proposed an action — please confirm or dismiss it above."the AI only produced a suggestion card and no textclick ✓ Execute or Dismiss on the card
The answer was there and is gone after StopStop only hides the display; the request continues on the serveropen the conversation again from History — the complete answer may be there
The AI no longer remembers earlier parts of a long chatonly the most recent 16 messages (about 32,000 characters) are sent to the AI, although the history keeps everythingrepeat the important facts, or start a new chat with a short summary
The AI stops after a few lookups or does not finish a multi-step taskper answer the AI may look things up in at most 5 rounds (each up to 120 seconds, the whole answer at most 5 minutes); lookup results are cut at 8,000 characterssplit the task into smaller questions

Display

MessageCauseSolution
Images appear as the link "view external image"deliberate: images are not loaded automatically, so that no image can report data to the outside world unaskedclick the link to open the image in a new tab
"Diagram could not be rendered:"the model delivered faulty diagram code; the code is shown belowask the AI to correct the diagram
"Mermaid could not be loaded:"the diagram library could not be loadedreload the page
Tool steps show English labels such as "Jira search: …", "Reading issue …", "Web search: …"these status rows are always English

Messages for actions in Jira and Confluence

Most of these messages appear on a suggestion card after ✓ Execute; a few, marked as such, come from the AI's own lookups and reach you only in its answer. Success messages are partly German even in the English interface: "✅ Vorgang ABC-1 erstellt.", "✅ Vorgang ABC-1 aktualisiert.", "✅ Comment added to ABC-1.", "✅ Page "…" created in space …" — each followed by a link.

MessageCauseSolution
"Write actions are not enabled by the admin."AI access is not Read + write — also when the administration changed it while the card was openask the administration to enable Read + writeGeneral Settings
"This feature has been disabled by your administrator."the surface was switched off meanwhileuse another entry point
"Required field missing: …"a required field is missing in the proposalask the AI to repeat the proposal with that field
"Field too long: …"a field exceeds 32,000 charactershave a shorter text generated
"Invalid field: …" / "Invalid arguments." / "Unknown action."the proposal is malformedask the AI to propose again
"Nothing to update: provide a new summary and/or description."an Update Jira issue card without a new summary or descriptionask the AI for concrete changes
"No access to Jira-Suche (user permissions insufficient)." / "No access to Vorgang ABC-1 (user permissions insufficient)." / "No access to Seite erstellen (user permissions insufficient)."you lack the rights in Jira or Confluenceclarify the rights in the project or space
"Vorgang ABC-1 was not found." (on a card; for lookups the AI may also report "Seite 123 was not found.")the issue or page does not exist or is invisible to youcheck the key; the AI can list the projects and spaces visible to you
"Space "DOCS" was not found."the space key is wrongstate the correct key
"Seite erstellen: Atlassian API responded with HTTP 400."Confluence rejected the page — most often a page with this title already exists in the spacechoose a different title
"Vorgang erstellen: Atlassian API responded with HTTP …" / "Kommentar auf ABC-1: Atlassian API responded with HTTP …" / "Vorgang ABC-1: Atlassian API responded with HTTP …"Jira rejected the action, e.g. because of required fields or an unknown issue type in that projectcheck project and issue type; ask the AI to adjust the proposal
"Invalid issue key." (on a card; for lookups the AI may also report "Invalid page ID.")wrong formatstate the correct key or page ID
"Action failed." / "Failed." / "Call failed."an unexpected error during executiontry again; if it recurs, inform the administration
The AI says it cannot find a page that was just createdfresh pages are sometimes not in the search yetwait briefly or name the page directly
After Update Jira issue images, @-mentions, status lozenges or smart links are gone; the description contains [Bild/Anhang] or [gekürzt, … Zeichen gesamt]the update replaces the whole description; the AI only sees images as the marker [Bild/Anhang] and text beyond about 8,000 characters is cutcheck the issue and restore what is missing from the issue history

Messages for history and archiving

MessageCauseSolution
"History could not be loaded."the history list was unavailablereload the page
"Conversation could not be loaded." / "Conversation not found."the conversation was deleted or dropped automatically — 50 conversations are kept per person, the oldest unpinned ones go firstpin or archive the important ones
"Fork failed."the copy could not be createdtry again
"No personal Confluence space found — open your Confluence profile once ("Create personal space") and try again."you have no personal Confluence spaceopen your Confluence profile, create the personal space, archive again
"Archiving failed." / "A page with this title already exists."Confluence unreachable, rights missing, or a title conflicttry again; if it persists, administration
Deletion with Save to my Confluence archive first does nothingarchiving failed — then nothing is deletedfix the archiving cause first, then delete
A conversation has disappeareddeleted, temporary, or dropped out of the most recent 50not recoverable; prevention: pin or archive
Older messages are missing in a long chatvery long conversations are trimmed at the front (the last two messages always stay)export or archive important content early
The archive page shows **bold** or tables literallyarchive pages are plain text by designuse ↓ Export conversation (Markdown) for a formatted copy
The archive page is in English although the interface is Germanarchive labels follow the Interface language set by the administration — German only if it is set to German (Deutsch)administration: General Settings

Messages for your own agents and skills

MessageCauseSolution
"At most 5 own agents."upper limit reachededit or delete an existing agent
"At most 10 own skills."upper limit reachededit or delete an existing skill
"The shared registry is full (max 30)."30 agents are already shared site-wide. Your agent has been saved as a private agentreopen it, clear Make available to all users of this site, save again
"„…“: name missing or longer than 40 characters." / "„…“: instructions missing." / "„…“: instructions longer than 8000 characters." / "„…“: description longer than 200 characters." / "„…“: emoji longer than 4 characters."a required field is empty or a limit is exceededfill in or shorten the field
"„…“: name missing or longer than 60 characters." / "„…“: text missing." / "„…“: text longer than 4000 characters."skill name or text empty or too longfill in or shorten
"Memory is limited to 4000 characters."the note is too longshorten it
"Unknown agent for this skill."the assigned agent no longer exists or is not available to youchoose a different agent or No agent
"Save failed." / "Speichern fehlgeschlagen: …" (this German wording also appears in the English interface) / "Delete failed."the request failedtry again
The agent or skill does not appear in the other productyour own agents, skills, memory and history are the same in Jira and Confluence — if something is missing, the product is not connected to the app or the configuration could not be loadedreload the page or ask your administration; do not create the agent again, it would count against your limit of 5
A placeholder such as {{Projét}} is not asked forplaceholders allow only letters A–Z/a–z, the umlauts ä ö ü Ä Ö Ü, ß, digits, spaces, full stops, hyphens and underscores (up to 40 characters)rename the placeholder

Messages in the administration

Saving and validation

These English messages appear next to the Save button when a save is rejected. Save on Modules and General saves the whole configuration, including connection cards edited but not yet saved on Connections.

MessageCauseSolution
"Save failed."validation or saving failed; the exact reason followscheck the entries
"Configuration could not be loaded: …" (red alert above the tabs)the configuration could not be readreload the page
"Add at least one connection."no connection has been createdcreate at least one connection
"„…“: name missing or longer than 40 characters."connection name empty or too longshorten the name
"„…“: endpoint must use https://." / "„…“: endpoint is not a valid URL." / ""…": endpoint must be a valid https URL."the address is faultyenter the full https address, e.g. https://llm.my-company.com/v1
"„…“: specify at least one model." / "„…“: at most 20 models." / "„…“: invalid model name „…“"no model, too many models, or a name with invalid characters (allowed: letters, digits, . - _ / :, up to 100 characters)correct the model list
""…": API key seems too short (at least 8 characters)."the key entered is too shortpaste the complete key
"Default model must belong to one of the connections."the default model refers to a removed connectionchoose a model from a saved connection
"Site knowledge must be text with at most 4000 characters."the text is too longshorten it
"At most 8 agents." / "„…“: instructions longer than 8000 characters."site agents: limit reached or instructions too longremove an agent or shorten the instructions
"At most 30 prompts." / "„…“: text longer than 4000 characters." / ""…": skill references an unknown agent — save the agent first."templates: limit reached, text too long, or the skill points to an agent that is not saved yetremove a template, shorten it, or save the agent on Agents first
"Web search: endpoint must use https://." / "Web search: an endpoint is required to enable it." / "Web search key seems too short (at least 8 characters)."web search settings incompletecomplete the address or key — Enabling Web Search and Image Generation
"API key deleted — connection is disabled until a new key is saved."you confirmed Really delete?enter a new key and save when needed
"Delete failed." / "Connection not found."the key could not be deletedreload and try again

Endpoint approvals

MessageCauseSolution
"Network approval for … was declined or failed — nothing was saved." (or "… (code …) — nothing was saved.")the Atlassian confirmation dialog was declined or cancelled while savingsave again and confirm the dialog
"These endpoint domains are configured but not yet approved for this installation: …" (yellow notice on every settings page)the approval is missing — typical after importing a configurationclick Approve now and confirm the dialog
"Approval was declined or failed — these domains stay blocked. Your configuration is unchanged."the dialog was declinedtry again
"Customer-managed egress is not available for this installation — these domains stay blocked until it is enabled." / "Saved, but the network approval could not be requested — …"Atlassian's approval function is not available for this installation; the configuration is saved but every address stays blockedcontact the vendor maplee via your administration
"Requests to … were blocked in … although the address appears approved. …" and cells blocked <date> in Approval status per productthe administrator's consent is missing in this product — the domain list is shown installation-wide, but the consent counts per productclick Re-approve in this product and confirm the dialog. A successful Test connection in this product also clears the entry; ordinary chat requests do not
"Only Jira administrators may change endpoint approvals. Your account has no Jira admin permission, so the actions on this page are disabled."you opened the Confluence settings page without Jira admin rightsask a Jira administrator, or open the settings in Jira
"Your Jira admin permission could not be verified from Confluence — the actions stay enabled. …"the check failed; the actions remain usableproceed, or use the Jira settings page
"The limit of 10 approved domains is reached. …" / "Atlassian allows at most 10 approved domains per installation. …"at most 10 approved domains per installationRevoke an unused address or use Replace
"Only https is allowed — API keys travel on this connection." / "Not a valid URL." / "This domain is already approved."the address entered is invalid or already approvedcorrect the address
"Revoking failed — the approval is still in place." / "Re-approving failed or was declined — nothing changed."the action was declined or failedtry again
Endpoint approvals with the notice 'Re-approve in this product' and the table 'Approval status per product' showing a blocked cell for Jira.

Connection test

MessageCauseSolution
"Connection ok — … model(s) at the endpoint." / "Connection ok." with "Visible for this key: …"the test succeeded
"Not in the list (may still work): …" (shown in red although the test succeeded)a configured model name was not reported by the servicenot necessarily an error — test the model in the chat
"Test failed." followed by the reasonsee the following rows
"Endpoint rejects the API key (401/403)."key wrong or expiredrenew the key
"Endpoint responded with HTTP …"the service reports an errorcheck address, provider type and quota
"Timeout after 10s."the service did not answer in timecheck reachability
"Connection failed: … This endpoint needs a network approval: check that an administrator has approved it under Settings → Endpoint approvals."the address is not reachable or not approvedManaging Endpoint Approvals
"The request was blocked inside Atlassian before reaching … — the endpoint is not approved for … yet. …"the approval is missing in this product; the test records a blocked entryapprove or Re-approve in this product
"Connection not found — please save first." / "No API key saved — please set a key and save." / "New — test available after saving."the connection is not saved yet or has no keysave first, then test

Messages for web search and image generation

MessageCauseSolution
"Web search is not enabled."web search is switched offadministration: Enabling Web Search and Image Generation
"Web search has no endpoint configured." / "Web search has no API key configured."the checkbox is ticked but address or key is missingadministration completes the settings
"Web search rejects the API key (401/403)."the key is wrong, expired or was discarded by a service or endpoint switchenter a new key
"Web search rate limit reached (429)."the search service's quota is exhaustedcheck the quota with the provider
"Web search timed out after 30s."the search service did not answer in timecheck reachability and the approval; with SearXNG also the JSON output format
"Web search failed: …" with a note about a network approvalthe search address is not approvedManaging Endpoint Approvals
The AI never searches the webthe tool is not offered because web search is off or address/key is missingadministration
"Image generation is not enabled." / "Image connection not configured."image generation is off or no image model is selectedadministration
"The configured image connection type does not support image generation." / the AI never generates imagesthe image model belongs to an Anthropic connection — it is then silently not offereduse an image model on an OpenAI-compatible or Azure OpenAI connection
"Image endpoint responded with HTTP …"the service reports an errorcheck model name and quota
"Image endpoint returned no URL (b64-only is not supported)."the service delivers the image in an unsupported formchoose a different image model
An image link no longer worksprovider image links are often valid only brieflygenerate the image again and download it promptly

Messages in the dashboard gadget

MessageCauseSolution
"Not configured yet — add a question via "Edit"."no question savedopen the gadget's edit mode and enter a question
"Please enter a question."the field was empty on Saveenter a question
"Save failed."the gadget configuration could not be savedtry again
"No answer yet — click "↻ Refresh"."no answer has been generatedclick ↻ Refresh
"Gadget could not be loaded."the app configuration was unavailablereload the dashboard
"Refresh failed." / "Error."the request could not be started or ended with an unspecified errorclick ↻ Refresh again; if it recurs, administration
"No response from stream — please try again."no data arrived for 60 secondsrefresh again
"Streaming currently unavailable."the gadget can only streamtry again later
"Hourly limit reached (… requests/hour) — please try again later."the personal limit of the person who clicked ↻ Refreshwait
"Not configured yet — …" / "The AnyLLM subscription for this site is inactive …" / "This feature has been disabled by your administrator."administration neededsee The app does not start or is locked

Common misconfigurations and recovery

SymptomLikely causeRemedy
Users see Not set up yet although connections existno connection has a stored API keyenter the key and save — Setting Up LLM Connections
The chat works in Jira but not in Confluencethe endpoint is not approved in Confluence, or Confluence is not connected to the appapprove the endpoint on the Confluence settings page; if the app is missing in Confluence, add Confluence to the installation in Atlassian administration (Jira is the required product; Confluence is added afterwards)
After importing a configuration every request failsthe network approval for the imported address is missingclick Approve now in the yellow notice — Managing Endpoint Approvals
A model is not foundthe model name differs from the service'sTest connection shows the visible model names
Users get the first model instead of the default modelthe default model belongs to a connection without a key or was removedchoose a default model whose connection has a key — General Settings
Answers are noticeably slowa model with an internal reasoning phasechoose a different default model
Answers break off without any textthe response budget is too small for a reasoning modelusers: raise Max response tokens or click Reset in ⚙ Model parameters …
Usage stays emptythe AI service returns no token usage in its streaming responsesask the provider whether its streaming API returns usage — Usage Statistics, Import/Export and Backup
The web search does not workthe address is not approved, the key is missing, or the key was discarded by a service or endpoint switchEnabling Web Search and Image Generation
Image generation is never offeredthe image model is on an Anthropic connectionmove it to an OpenAI-compatible or Azure OpenAI connection
Users still see disabled surfaces in the menuplatform behaviour — menu entries cannot be hidden per installationexpected; see Modules
The chat bar stays visible or hidden for a while after a changeits status is cached in the browser for 10 minuteswait or clear the site data
The settings cannot be openedmissing administration rightscheck the rights in the product
The approval actions are disabled on the Confluence settings pageno Jira admin permissionuse a Jira administrator account

What you can get back if in doubt

  • Configuration: create a ↓ Export as JSON regularly (Import/Export). It contains everything except the API keys — Usage Statistics, Import/Export and Backup.
  • API keys: cannot be read out. Keep them where you otherwise manage them.
  • Conversations and memory: sit with the individual people and can neither be backed up nor restored centrally. Anyone who wants to keep a chat permanently archives it to Confluence — Managing Chat History.

Contacting support

Turn to your own product administration first — they can view connections, approvals and settings. If the administration has to escalate further, the vendor of the app is maplee; ask your administration for the agreed support routes. This manual deliberately names no addresses or response times.

This information helps — and is safe to share:

  • the product (Jira or Confluence) and exactly where you were (e.g. "issue panel", "chat bar", "dashboard gadget")
  • the message verbatim
  • the time, with time zone
  • what you did before it occurred
  • whether it is reproducible and whether it affects others too
  • whether it occurs in both products

Please never include: API keys, passwords, tokens, complete chat content with confidential data, or personal data of third parties. For diagnosis, the message, the time and the surface are enough.

Rendered from the app’s own interface with sample data; the Jira/Confluence frame around it is not shown.

Documentation baseline: app version 0.2.0 · 2026-08-30