602 signals · 626 observation events
The open source coding agent.
v1.18.18Prereleases and milestones indicate public plans; they are not delivery commitments.
### Description qwen3.6 27b model updates to agents.md always trigger prompt processing. I run the model using llama.cpp, and opencode works fine at any time, rarely triggering prompt processing to re-read the context. However, as soon as the agent updates agents.md, it always writes a small piece of content to trigger one prompt processing. At 200k context,
## Description The MiniMax CN provider (`api.minimaxi.com`) fails with a certificate error in the desktop app but works perfectly in the TUI. **Error:** `Cannot connect to API: certificate has expired` ## Environment - **OS:** Windows 11 - **OpenCode version:** 1.17.6 - **Provider:** `minimax-cn` (`https://api.minimaxi.com/anthropic/v1`) - **Works in:** TUI
### Description [chris@worksoul:/vmTmp/git-repos/DevDashboard/1UGMsy6K]$ opencode --log-level DEBUG debug config Error: Unexpected error disk I/O error [chris@worksoul:/vmTmp/git-repos/DevDashboard/1UGMsy6K]$ opencode --version 1.17.7 ### Plugins _No response_ ### OpenCode version 1.17.7 ### Steps to reproduce _No response_ ### Screenshot and/or share link _
### Description Users with an active OpenCode Zen balance are still being limited as if they are free-tier users. I tested this with two separate users/accounts. One account had a Zen balance of at least $20, but both accounts still hit the same free usage cap / 200 request limit and received 429 rate-limit errors. The paid/balance-enabled account did not ap
### Description UPDATE: There might actually just not be a backend running because sending a prompt hung and didn't even give a spinner. The opencode TUI crashed with an unexpected error. **Error:** opencode server GET http://127.0.0.1:4096/vcs/diff?mode=git&context=12&directory=%2Fhome%2Fjohn%2F.config%2Fwaybar → 404 Not Found: (empty response body) **Stack
### Description The `anomalyco/opencode/github` Action's `agent` input is silently ignored — the model always responds using the default (`build`) agent instead of the one configured. Confirmed this is **not** a general agent-resolution bug: `opencode run --agent <name>` (same `opencode` version, run manually) applies the agent correctly. The bug appears iso
### Description When OpenCode runs as an ACP agent in Zed, the built-in `/compact` command is not included in `available_commands_update`. Zed rejects it as an unrecognized command before it reaches OpenCode, even though the ACP service already handles `/compact` through `session.summarize()`. ### Plugins None required. ### OpenCode version 1.18.2 ### Steps
i started working with it but when i came back for some reason it didnt work again ...i cant seem to get answers its showing thinking then it stops abruptly ...ive tried several things even tried new sessions same issue
我订阅了opencodego怎么又用不了了
my app was working great and all of a sudden i get no response it makes a clicking noise and just where i submit my text i tried uninstalling and re downloading now it does the same thing just with a interruption tag with no other response
### Verify - [x] I have verified this feature I am requesting has not been suggested before. ### Describe the enhancement you want to request `packages/docs/` is a leftover Mintlify starter template (README says "Mintlify Starter Kit", quickstart.mdx/development.mdx/index.mdx and essentials/ are all generic Mintlify onboarding content, not OpenCode docs). It
### Description I have a project lets say its called `projects/foo` and i have `projects/foo2` now opening `foo` or `foo2` will always open `foo` and there is seemingly no way to open the `foo2` project. ### Plugins none ### OpenCode version v1.16.2 ### Steps to reproduce create 2 project folders, `foo` `foo2` copy content from `foo` to `foo2` open `foo` in
Hello! I'm having a problem when using DeepSeek V4 Pro. I get the following error message: "The latest version of this model is only available hosted in China and requires explicit opt-in." Could you please help me understand how to enable it or resolve this issue?
Env: Windows 11, Codex with `wire_api = responses` via cc-switch → `https://opencode.ai/zen/go/v1`, model `deepseek-v4-pro`. I retried right after the 0813 model update and the problem is still there. **What happens**: First turn in a fresh conversation is fine. From the second turn on, as soon as the request includes the previous assistant reply, it returns
### Description Hello. Moonshot AI can not be used for now. `Invalid request: the message at position 5 with role 'assistant' must not be empty` all Kimi K2 models shows same error. ### OpenCode version 1.0.193 ### Steps to reproduce _No response_ ### Screenshot and/or share link _No response_ ### Operating System _No response_ ### Terminal _No response_
### Description ## Description I am reporting a persistent quota issue with my OpenCode Go subscription. Despite having **zero usage since April 23**, my account is still flagged as "Subscription quota exceeded" as of April 26. There is a significant discrepancy between my actual usage history and the dashboard status, preventing me from using the service fo
### Description ## Summary I use `deepseek-v4-flash` for Hermes agents and have no `glm-5` configured. But the Usage page shows both `glm-5` and `deepseek-v4-flash` being used. ## Expected Only configured models should be shown, and the Usage page should show the exact model used, not unused models. ## Actual The Usage page shows `glm-5`, and it is not clear
### Description OpenCode's ACP support, tested through Zed, after installing OpenCode as an external agent, doesn't broadcast how big the context is, and how full it currently is, whereas other agents do. I'm not an expert in ACP, but I think this is what it should send: https://agentclientprotocol.com/rfds/session-usage#session-context-size-and-cost example
### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request Description: Add a slim sidebar on the right side of the TUI that lists only the user's own inputs (prompts) from the current conversation as clickable entries. Clicking an entry jump
### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request I would like to open the existing TUI diff viewer directly from the CLI. Example: ```bash opencode diff opencode diff . opencode diff /path/project ``` Currently i need to start openc
### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request Grok 4.3 in OpenCode currently has no reasoning levels. Grok 4.3 on API has none/low/medium/high, hopefully this would work with OAuth as well Thanks!
### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request When using layered configuration files (e.g., a global configuration in `~/.config/opencode/` and a local project configuration in `.opencode/opencode.json`), defining `skills` in a h
### Description I was in plan mode working, and my agent ran `gh issue comment` and posted a reply directly on the issue which I'm working on — without being asked to, without switching to build mode, and without triggering any permission prompt. ### Root cause The plan mode system prompt explicitly says: > Do NOT use sed, tee, echo, cat, or ANY other bash c
## Problem I subscribed to OpenCode Go today (payment completed, first month 50% off, $5 charged). After subscribing, I still receive: Free usage exceeded, subscribe to Go retrying in 32969s - attempt #1 The system still treats my account as free-tier. The subscription appears to have no effect. ## What I already tried - Verified the payment went through (ca
### Description When an HTTP provider connection resets, OpenCode aborts the active session drain. If the reset happens after response headers and valid SSE events have arrived, the incomplete assistant step fails and there is no recovery beyond manually prompting the session again. Concurrent child sessions also lose their active work when their streams res
### Description After several successful tool calls suddenly one toolcall ends and opencode TUI remains stuck in busy (ping pong animation ongoing). Double ESC to interrupt does nothing. The last log message is either "level=INFO run=XXXX message=loop session.id=ses_XXXX step=N" or something like "level=INFO run=XXXX message=evaluated permission=bash pattern
### Description Merge Gateway models can declare supported reasoning efforts through models.dev, but OpenCode does not map `merge-gateway-ai-sdk-provider` to the adapter's `mergeGateway` provider-options namespace. As a result, the declared effort variants are unavailable and a selected effort is not forwarded to the adapter. Expected behavior: expose the ex
## Description Sessions accessed via the prompt API from another session are silently deleted. This happens consistently after a session is accessed through cross-session API calls. ## OpenCode version 1.18.16 ## Steps to reproduce 1. Create two sessions (A and B) in OpenCode web UI 2. From session A, send a prompt to session B via the API: ``` POST /session
### Description Two typos in the docs: 1. `packages/web/src/content/docs/providers.mdx:2490` and `:2528` — `"name": "My AI ProviderDisplay Name"` is missing a space; should be `"My AI Provider Display Name"`. 2. `packages/web/src/content/docs/github.mdx:100` — "a [personal access tokens]" uses the plural form after "a"; should be "a [personal access token]".
As the picture, build is success and all the tasks were done, but agent keep thinking forever and I need to stop it manual. <img width="1920" height="1020" alt="Image" src="https://github.com/user-attachments/assets/ea47b346-6d81-4e1e-a209-7a9df9421173" />
### What happened A key bound to `app.toggle.session_directory_filter` does nothing while the session picker is open. The same key works from the main screen. The default for this keybind is `none`, so only users who bind it hit this. ### Steps to reproduce 1. Put this in `tui.json`: ```json { "keybinds": { "app_toggle_session_directory_filter": "ctrl+a" } }
### Description `apply_patch` deletes the final blank line of any file whose content ends with `\n\n`, even when the patch doesn't touch that line. The tool's own response confirms the unrequested change: it reports 2 deletions for a patch that asked for 1, and repeated edits keep eating one trailing newline each time. Root cause: `derive()` in `packages/cor
## Summary On the OpenCode Zen gateway `https://opencode.ai/zen/go`, the Anthropic-compatible endpoint `POST /v1/messages` fails for **`deepseek-v4-pro`** when `messages[].content` is passed as a **string**. The exact same request with **`deepseek-v4-flash`** returns 200 OK. Switching `deepseek-v4-pro` to the **array form** of `content` also works. This only
Description: I exceeded the free usage limit during an active session. After subscribing to Go and adding my token, new chats work perfectly under the subscription. However, the existing conversation remains permanently wedged in a retry loop showing "Free usage exceeded, subscribe to Go" and will not recognize the updated subscription state. Steps to reprod
I puchrased the subscription yet the chat still says limit exeeded buy Go. Also when trying to use Deepseek for Go it says its only available in china, it was my mane engine.
## Bug description When using the native Azure OpenAI provider (`azure`) in OpenCode, small models like `gpt-5-mini` work correctly, but larger models (`gpt-5.6-luna`, `gpt-5.6-sol`, `gpt-5.4`, `o3`) hang indefinitely instead of returning a response. ## Steps to reproduce 1. Configure Azure OpenAI provider with a valid resource and API key: ```powershell $en
## Summary When the terminal is resized, the opencode TUI re-layouts on **grow** but not on **shrink**, and not on the initial attach. The drawn width stays at its previous value, so the TUI either leaves blank space or overflows the terminal. Measured in a browser-hosted terminal (xterm.js over a WebSocket PTY, opencode 1.18.16): | terminal columns | column
### Reproduction ```sh docker run -d --name opencode-eof node:22-bookworm-slim sleep infinity docker exec opencode-eof npm install -g opencode-ai@1.18.18 timeout -s KILL 8s docker exec opencode-eof opencode </dev/null docker top opencode-eof -eo pid,ppid,tty,stat,etime,args ``` ### Actual `opencode` does not exit. After the `docker exec` client is killed, th
### Description 我认为此问题应该是其他问题的重复 因为我之前有遇到过此问题，于是我更新了结果更新之后又遇到该问题 正常情况下，他应该让我等到明天早上再使用opencode但他并没有他一直在尝试重新尝试 如果您认为我冒犯了您，您随时可以关闭这个 I think this question should be a duplicate of other questions Because I have encountered this problem before, I updated the results and encountered the problem again. Normally he would have asked me to wait until tomorrow morning
### Description I use Routerly and OpenCode, and have been using a combination of OpenCode Zen and Go subscriptions. Until today, this worked fine until I noticed models running very slow - timeouts. To help work around those issues, I re-added the models and created policy in an attempt to see if Routerly will choose models that are performing better. (Perf
### Description The [tools docs](https://opencode.ai/docs/tools/) say `websearch` "is only available when using the OpenCode provider or when the `OPENCODE_ENABLE_EXA` environment variable is set". I'm using the official **OpenCode Go** subscription ([docs](https://opencode.ai/docs/go/), model IDs use the `opencode-go/` prefix, e.g. `opencode-go/deepseek-v4-
## What's New ### 🚀 Workspace Selection for New Sessions - **Start sessions where you want**: Choose between **Local repository**, spin up an isolated **New workspace**, or select from **Existing workspaces**. - **Context-aware composer pill**: The location selector shows branch context, respects project defaults (*Last used*, *Local*, or *New*), and include
## Summary When an `opencode.json` contains a `{file:./path}` reference that doesn't exist on disk, OpenCode rejects the **entire** agent configuration with a hard error: ``` Error: Configuration is invalid at /var/www/extrachill.com/opencode.json: bad file reference: "{file:./AGENTS.md}" /var/www/extrachill.com/AGENTS.md does not exist ``` This means no age
### Description A subagent called at depth 3 (parent → child → grandchild) fails silently without any error message. The issue does not occur at depth 2 (parent → child). The problem began appearing on 2026-04-17, while everything worked normally the previous day. The grandchild agent receives the prompt successfully (child session window appears), enters a
### Description I'm experiencing an error when trying to open files/projects using the tabs/buttons in the OpenCode desktop app. The operation fails with the following error: Copy Request failed Error invoking remote method 'open-path': Error: spawn [object Promise] ENOENT Environment Details: App: OpenCode Desktop Application Location: ~\Desktop\game\cleani
### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request **OpenCode doesn't support RTL (Right-to-Left) text direction for Arabic or Persian.** When users write in Arabic, the chat display shows text incorrectly, making it hard to read and
### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request Add an optional `version` input to `anomalyco/opencode/github`. The install script (`https://opencode.ai/install`) already supports `--version` to install a specific release: ``` curl
### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request Add SHA256 checksum verification to the install script (`install`). The script downloads binary archives from GitHub Releases and extracts them, but performs no integrity check after
### Description Deepseek V4 Flash Free doesn't consistently broadcast reasoning effort: When I open zed, External agents, Opencode, I have it set to Deepseek V4 Flash Free. However, initially, I don't see the reasoning level select. I have to select a different model and then select Deepseek back to see it. Before: <img width="781" height="181" alt="Image" s
## Bug Description When copying text containing CJK characters (Chinese, Japanese, Korean) from opencode running in VSCode integrated terminal on WSL2, the pasted result is garbled/mojibake. ASCII text copies correctly. This is a known issue in the upstream OpenTUI framework's OSC52 clipboard implementation. The same bug was reported and fixed in charmbracel