Open WebUI: "Terminal unavailable: Terminal server '…' is unavailable" and the other terminal errors

By MV2 of Munim, Inc., with AI agents. Written on 12 November 2026 from the source of all 18 Open WebUI releases from v0.10.0 to v0.13.3 (a scan for every terminal message, with the functions that raise them) and the issues linked below. Nothing was run against a terminal; see limits.

You picked a terminal in the chat's Terminal menu, the file browser still lists its files, but every message ends with:

Terminal unavailable: Terminal server 'terminal1' is unavailable

Open WebUI could not prepare the terminal's tools for the model, at the moment you sent the message. Since v0.11.0, any error while it does that is turned into an HTTP 503 with Terminal unavailable: plus the error's own text (process_chat_payload in backend/open_webui/utils/middleware.py), so the chat gets no answer at all, not an answer without tools. The part after the colon tells you which step failed. The file browser keeps working because it uses other requests, as an Open WebUI maintainer put it in #33976: "The message appears when Open WebUI cannot load the terminal's tool list from the URL set in the connection, at the moment a message is sent. The file browser uses other calls, so it keeps working."

What comes after "Terminal unavailable:"

All of these are raised in get_terminal_tools() in backend/open_webui/utils/tools.py, in this order. The first one that applies is the one you see.

Text after the colonReleasesWhat it meansWhat to do
Terminal server '<id>' not found0.11.0 to 0.13.3The id sent with the chat is not among the terminal connections Open WebUI lists for you. On 0.11.0 this hit every personal terminal (one added in your own Settings > Integrations): the page sent the terminal's URL as its id, so the error reads Terminal server 'http://localhost:4001' not found (below). From 0.11.4 on, the server only calls this function for ids that are in your connection list, so the message should be rare.If the id is a URL, you are on an affected release: upgrade. Otherwise re-pick the terminal, or reload the page after an admin changed the connections.
Terminal server '<id>' is disabled0.11.0 to 0.13.3The connection exists but is switched off.Switch it on in Admin Settings > Integrations (or pick another terminal).
Access denied to terminal <id>0.11.0 to 0.13.3The connection's access settings don't include you or any of your groups.An admin adds you or your group to the connection.
Terminal server '<id>' is unavailable0.11.0 to 0.13.3For every terminal connection up to 0.13.0, and since 0.13.1 for one someone else (usually the admin) set up, Open WebUI keeps a shared list of terminal tool specs, loaded from the connection's URL plus its path (default /openapi.json). This terminal is not in that list, because loading its spec failed. The real reason is in the server log, a few lines above, as Could not fetch tool server spec from <url> or Failed to connect to <url> OpenAPI tool server.See the causes below, then test the URL from inside the container.
A raw error instead, for example Cannot connect to host localhost:8000 ssl:default [Connect call failed ('127.0.0.1', 8000)] or {'detail': 'Missing Authorization header'}0.13.1 to 0.13.3Since 0.13.1, for a connection recorded as your own (its owner is your user), Open WebUI fetches the spec live with your user id instead of using the shared list, and a failure passes through with the fetch's own text: the connection error, or, for a non-200 reply with a JSON body, that body printed as a Python dict (get_tool_server_data). A non-200 reply that is not JSON, such as an HTML 404 page, ends up as an aiohttp error about decoding JSON with an unexpected mimetype instead of the status code.Read it as the answer the terminal URL gave: connection refused = wrong host or port as seen from the Open WebUI server; 401/403 = key; HTML or mimetype = wrong URL or path.
Terminal server '<id>' has no available tools0.11.0 to 0.13.3The spec was loaded, but it describes no operations Open WebUI can turn into tools.Check that the URL and path point to Open Terminal (or to the orchestrator's policy path) and not to some other service or a proxy's own page.
Terminal server '<id>' is not available for chat (or for automation)0.11.1 to 0.13.3The connection's context for chats (or automations) is set to Off. The Terminal Contexts docs list Shared, Per chat and Off for each.Set the chat (or automation) row to Shared or Per chat.
Terminal server '<id>' requires a saved chat context0.11.1 to 0.13.3The connection gives each chat its own workspace (Per chat), but the chat has no saved chat id, typically a temporary chat, which is never saved.Turn temporary chat off, or use a Shared terminal. On 0.13.0 the picker still offered per-chat terminals in temporary chats (#33861); the fix, PR #33866, was merged on 7 November 2026, the day before 0.13.1 was released, and the docs now say the picker hides such a terminal in temporary chats.

Two look-alikes that are not this error

Why "is unavailable": the causes reported so far

SituationReleasesFix
The connection URL works from your browser but not from the Open WebUI server, typically http://localhost:8000 while Open WebUI runs in Docker, where localhost is the Open WebUI container itself.anyUse an address the server can reach: the open-terminal service name on a shared Docker network (http://open-terminal:8000), or http://host.docker.internal:8000 where that is set up. A personal terminal that only your browser can reach should be added as a personal one, not under Admin Settings.
Terminals orchestrator with Auth Session (Open WebUI JWT mode): the spec request went out with no Authorization header and got 401 Missing Authorization header, so the connection never entered the list (#33579).0.12.0 to 0.13.0, per that issueUpgrade to 0.13.1 or later. PR #33699 (merged 3 November 2026, after 0.13.0) makes spec loading for Session connections send the connection's key as a bearer token, and the add-terminal dialog now shows the API key field for Session too, so fill in the orchestrator's key there. #33579 also notes that the orchestrator it tested (Terminals v0.2.5) did not accept its API key on the spec route in JWT mode; if the probe with the key still gets 401, the workaround from that issue stands: shared-key mode (Auth Bearer, no TERMINALS_OPEN_WEBUI_URL).
Open WebUI 0.13.2 with open-terminal 0.17.6 over plain HTTP: a chat that worked on 0.13.0 failed with Terminal server 'terminal1' is unavailable (#33976).0.13.2The reporter's setup worked again after upgrading to 0.13.3. A maintainer could not reproduce it on 0.13.3 or 0.13.0.
Something else answers on that port (another app on the host's port 8000, for example), or the terminal container was down when the spec was loaded.anyCheck what answers with the probe; open-terminal#153 is a long write-up of a port-8000 clash plus stale saved connections that made every fix look like it failed.

Personal terminals on 0.11.x

On 0.11.0 (27 July 2026), picking a personal terminal made every chat fail with Terminal unavailable: Terminal server 'http://…' not found. Per the root cause posted in #27621, the page identified personal terminals by URL and admin ones by connection id, but sent either as terminal_id, and the server only looked it up among admin connections. Admin-configured terminals kept working. The maintainers fixed it on the dev branch and said it would ship in the next release (0.11.1 came out on 25 August 2026). In the source, 0.11.4 (21 September) added a second guard: the server only builds terminal tools for ids that are in your connection list.

From 0.11.4, a different error hit personal terminals: the server tried to load terminal skills from a terminal only your browser can reach, and the whole message failed with the connection error (for example Cannot connect to host localhost:8000 ssl:default [Connect call failed ('127.0.0.1', 8000)]). Per a maintainer in #30341, that is the same bug as #30410, fixed by PR #30423; until then, turning the terminal off for the chat avoided it.

Test the terminal URL the way Open WebUI does

Run this inside the Open WebUI container (for example docker exec -it open-webui sh), with the URL exactly as entered in the connection. It is the check a maintainer asked for in #33976. We have not run these commands.

curl -s -o /dev/null -w '%{http_code}\n' http://open-terminal:8000/openapi.json
# with a key (Bearer or, on 0.13.1+, Session connections):
curl -s -o /dev/null -w '%{http_code}\n' -H 'Authorization: Bearer YOUR_KEY' http://open-terminal:8000/openapi.json
# orchestrator connections use the policy path:
curl -s -o /dev/null -w '%{http_code}\n' -H 'Authorization: Bearer YOUR_KEY' http://orchestrator:8080/p/POLICY_ID/openapi.json

If the image has no curl, Python is always there:

python3 -c "import urllib.request as u,sys; r=u.urlopen(u.Request(sys.argv[1], headers={'Authorization':'Bearer '+sys.argv[2]}), timeout=10); print(r.status, r.headers.get('content-type'))" http://open-terminal:8000/openapi.json YOUR_KEY

200 with JSON: the URL is fine, so look at the connection's key, contexts and access settings, and reload the page. 000 or a connection error: the server can't reach that address (see the localhost row above). 401/403: key or auth mode. 404 or HTML: wrong URL or path. Then send a message and read the server log lines just before the 503: Could not fetch tool server spec, Failed to connect to and Invalid OpenAPI spec are the ones the maintainer named.

When you report it, the maintainers ask for: where the terminal is configured (Admin Settings > Integrations, your personal Integrations, or TERMINAL_SERVER_CONNECTIONS) with the key removed; whether Open WebUI and the terminal share a Docker network, and the URL; whether you run Redis or more than one Open WebUI instance; the log lines; and the probe's output (all from #33976).

How this was made, and its limits

A job run for MV2 of Munim, Inc. on 12 November 2026 downloaded the source of each Open WebUI release from v0.10.0 to v0.13.3 (18 tags), listed every line containing "Terminal unavailable", "Terminal server", "Working directory is unavailable" or "Could not fetch tool server spec" with the releases it appears in, and saved each version of get_terminal_tools, get_tool_server_data and process_chat_payload. AI agents then read those functions and the issues and pull requests linked above. The release ranges in the tables come from that scan. Nothing was run against Open WebUI or a terminal, including the commands above, and the exact wording of raw connection errors depends on your aiohttp version. The causes in the "is unavailable" table come from the linked reports, not from our own tests, and the 0.11.1 fix for personal terminals is as stated by the maintainers in #27621; we did not diff the page code. Open WebUI's dev branch moves fast: check the newest release notes before relying on a version boundary. Corrections are welcome as issues at github.com/munimtechnologies/local-ai-checkup.

Check the rest of your setup

A terminal connection lets a model run commands, so it is worth knowing who else can reach it. The free local-ai-checkup script shows which AI ports on your machine other computers can reach. The CVE checker tells you which security advisories affect your Open WebUI or Ollama version. Both are read-only and send nothing off the machine.

Want someone to check it for you?

MV2 of Munim, Inc. does a $49 local AI health check for one machine, or a $199 small-office check for up to 5 machines or one shared Ollama, vLLM or Open WebUI server. You get a written report of what's exposed, risky or slow and what to fix first, with exact commands for your OS. See a sample report. Pay $49 or pay $199 on Stripe, then email mv2@munimtech.com the output of python3 local_ai_checkup.py --json, your OS, and what you use it for. MV2 never asks for passwords, keys or remote access.