Skip to content

[tui] Assistant response never renders - blank reply area after sending message (v2.0.26) #54447

Open

Description

@sanirudh17

Description

In the opencode2 TUI (NOT the Desktop app), when I type a message and send it, it shows as working but the assistant response never renders — the reply area stays blank/empty (see screenshot). Both messages (Hello and What model are you?) show only the user bubble with no assistant reply visible, even though the bottom status bar shows Build · Step 5 Preview Free OpenCode Zen · high and the session appears active.

Plugins

  • local: orca-opencode-status-tui
  • local: orca-opencode2-status-tui

OpenCode version

v2.0.26 (opencode --version → opencode v2.0.26)

Steps to reproduce

  1. Open opencode2 TUI on Windows (opencode)
  2. Start a New session
  3. Type Hello and send — no assistant reply renders, area stays blank
  4. Type What model are you? and send — same: user message shows, no assistant response appears

Screenshot and/or share link

Screenshot attached in original report: user messages Hello and What model are you? visible in center pane, no assistant responses rendered below them; right pane shows New session - 2026-10-11T04:32:13.373Z; bottom bar shows Build · Step 5 Preview Free OpenCode Zen · high. Unable to run /share since no response completes.

Operating System

Windows 11 26H2 (Build 26300) — [System.Environment]::OSVersion = Microsoft Windows NT 10.0.26300.0, PowerShell 5.1.26100.9444. (Get-ComputerInfo misreports product name as "Windows 10 Home" on this build; reporter confirms Windows 11 26H2.)

Terminal

opencode2 TUI inside Windows Terminal (PowerShell 5.1 host); CLI version opencode v2.0.26. NOT the OpenCode Desktop app.

Activity

  1. sanirudh17 commented on Oct 11, 2026

    @sanirudh17
    Author

    Correction: this is the opencode2 TUI (running inside Windows Terminal / PowerShell), NOT the Desktop app. Updated title and body accordingly.

  2. changed the title [-][desktop] Assistant response never renders - blank reply area after sending message (v2.0.26)[/-] [+][tui] Assistant response never renders - blank reply area after sending message (v2.0.26)[/+] on Oct 11, 2026
  3. opencode-agent commented on Oct 11, 2026

    @opencode-agent
    Contributor

    Thanks for the details! I tried this on Linux with opencode v2.0.26 and the same model, Step 5 Preview Free (OpenCode Zen, high variant). Through the CLI I sent Hello and then What model are you?, and the model replied normally both times. The saved replies contain ordinary text parts, not empty or reasoning-only turns. So the model and provider seem fine, and the problem looks specific to the Desktop app on Windows, which I can't run here. A maintainer with that setup will need to look at it.

    These could help narrow it down:

    • In a terminal, does opencode run -m opencode/step-5-preview-free "Hello" print a reply on the same machine?
    • Do other models show replies in Desktop, or is it only Step 5 Preview Free?
    • Does it still happen with the two local Orca plugins turned off?
    • Does a notice like "The server restarted while you were working" ever appear, or does the session stay busy forever?
    • Any errors in the logs under %USERPROFILE%\.local\share\opencode\log\ from around the time you sent the message.

    Related issues, most likely first:

  4. sanirudh17 commented on Oct 11, 2026

    @sanirudh17
    Author

    Diagnostic findings (run on the reporter's machine)

    Context: reporter upgraded from opencode v1 → v2 via the built-in /update command (log shows message="upgraded major" method=npm ... @opencode/[email protected] at 2026-10-11T04:31:55Z, ~40s before the broken session was created at 04:32:13Z). Current version opencode v2.0.26, Windows 10 Home Build 26100, opencode2 TUI in Windows Terminal / PowerShell 5.1.

    1. opencode run -m opencode/step-5-preview-free "Hello" — FAILS, same symptom outside the TUI

    • Command hangs with zero stdout and hits the 30s timeout, twice (shared server AND --standalone).
    • So this is not a TUI render bug — the request itself never completes anywhere.

    2. Other models — also FAIL, not Step-5-specific

    • opencode run --standalone -m opencode/gpt-5-nano "Reply with the single word OK" → same 30s hang, no output.
    • opencode auth list → also hangs (30s timeout), suggesting a broader provider/auth stall, not one bad model.
    • Broken session (ses_ed6c469ccffekFPB32xCGfekrt) exported as messages: [], tokens: {input:0, output:0}, cost: 0 — the prompt never persisted server-side.
    • Note: an older pre-update session on 2026-10-10 shows a real upstream error: AI_APICallError: Upstream request failed: Insufficient account funds on gpt-5.4-nano — but post-update requests hang instead of erroring.

    3. Orca plugins OFF — still FAILS

    • Disabled TUI plugins (tui.jsonc → { "plugin": [] }, so neither herdr-tui-session.js nor orca-opencode-status-tui/tui.js load), restarted the background server (opencode service restart → http://127.0.0.1:49374 healthy), re-ran the Step-5 test → still hangs.
    • Full plugin dir for reference: orca-opencode-status-tui, orca-opencode2-status-tui, wt-agent-hooks, herdr-agent-state.js, orca-opencode-status.js, orca-opencode2-status.js, pebrel.js (+ .pebrel-managed, .js twins). Log also shows failed to load plugin wt-agent-hooks.js: must default export an object with server().
    • Restored tui.jsonc to the original two entries afterwards (verified).

    4. Server restart notice vs busy forever — STUCK, no restart notice

    • No "The server restarted while you were working" line anywhere. prompt-history.jsonl confirms the messages were typed (Hello, What model are you?, /btw DId you respond?), but:
    • POST /api/session/ses_ed6c469ccffekFPB32xCGfekrt/prompt → http.status=499 with InterruptError: All fibers interrupted ... at SessionPrompt.prepare — the request was abandoned server-side, no reply ever streamed. Newer sessions (ses_ed6bffdd8ffeopMEruGXcvunDm, ses_ed6bbeb4fffe52crmAOKdb5NEb) hit the identical 499 pattern.

    5. Log errors around message time (%USERPROFILE%\.local\share\opencode\log\opencode.log)

    • 04:32:08–04:32:11Z: server boot, plugin loads, TUI connects normally. One line of interest: database migration started/completed migration=20261007190000_azure_cli_external_credential right at boot post-update.
    • 04:36:49Z: POST /api/session/.../prompt → 499 InterruptError ... SessionPrompt.prepare — the reported messages dying here.
    • 04:37–04:41Z: same 499/interrupt pattern repeats for follow-up sessions and my opencode run repros.

    Migration hypothesis

    The v1→v2 /update path (npm 1.19.0 → @opencode/[email protected] within ~2 min, plus a v1-era package.json still pinning @opencode-ai/plugin 1.4.7 in the global config dir) looks like the prime suspect: immediately after the update, auth-gated paths (run, auth list) stall instead of erroring, while non-auth paths (models, session list/export, service status) work fine. Plugin interference is ruled out by test 3.

  5. sanirudh17 commented on Oct 11, 2026

    @sanirudh17
    Author

    OS correction: reporter is on Windows 11 26H2 (Build 26300). The earlier Windows 10 label came from Get-ComputerInfo misreporting the product name. Issue body updated.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions