602 signals · 626 observation events
The open source coding agent.
v1.18.18Prereleases and milestones indicate public plans; they are not delivery commitments.
### Issue for this PR Fix #7380 ### Type of change - [x] Bug fix - [ ] New feature - [ ] Refactor / code improvement - [ ] Documentation ### What does this PR do? 1. Add lazy-scroll loading: • Scrolling up: When user reaches within 5px of the top, loads next 50 older messages via before cursor • Scrolling down: When user reaches within 5px of the bottom, loa
### Description # [BUG] MCP tools connected but not exposed to agent ## Description MCP server `pdfrag` connects successfully and exposes 6 tools via `tools/list`, but the tools do not appear in the agent's available tool list. The server is working correctly (responds to MCP protocol, tools are valid), but something in the opencode pipeline isn't passing th
### Description When using OpenCode inside VSCode Server (Docker environment), copying text from an OpenCode session shows the message Copied to clipboard, but the text is not actually copied to the system clipboard and cannot be pasted elsewhere. ### Plugins None ### OpenCode version 1.18.14 ### Steps to reproduce 1. Start VSCode Server inside a Docker cont
## Summary Ingesting a large PDF — either by **pasting its path into the input box** (renders as a `[PDF N]` chip) or by **letting the agent read it with the `read` tool** — base64-encodes the **entire file** into memory with **no size limit**, and the resulting `data:application/pdf;base64,...` string is **duplicated on every subsequent turn**. The same bas
new vision（1.18.18.0）：Upstream request failed: [invalid_bearer_credential] Missing or invalid bearer credential
### Description <img width="606" height="377" alt="Image" src="https://github.com/user-attachments/assets/f5e4a0da-23f9-4e2d-8d07-2dc57dd16199" /> ### Plugins opencode ### OpenCode version 1.3.9 ### Steps to reproduce /connect 选择OpenCode Go 输入订阅的OpenCode的api秘钥 选择GLM-5 然后随便问问题就需要十多分钟才能响应出结果 ### Screenshot and/or share link _No response_ ### Operating System w
### Description gpt-5.5 isn't working, I have the Plus plan <img width="845" height="133" alt="Image" src="https://github.com/user-attachments/assets/77832a56-66be-43b8-82b8-cd4e10a1bb16" /> <img width="842" height="190" alt="Image" src="https://github.com/user-attachments/assets/c664991d-8568-40d1-93f3-cb3d1d78e03a" /> ### Plugins No ### OpenCode version la
### Description When I use minimax-m3 - opencode doesn't collapse all thinking chunks into one - each think is ~3-4ms. The final answer is not in the end of the message and might be somewhere in the middle. <img width="453" height="422" alt="Image" src="https://github.com/user-attachments/assets/24bfc092-cc10-4a71-83ec-eca129d8063c" /> Each chunk looks like
## Feature Request ### Checklist - [x] I have searched existing issues and confirmed this feature has not been suggested before (checked: #31932 and #17842 which are related but focus on CLI flags; this issue focuses on the TUI session picker interaction model) ### Problem The `/session` picker in the TUI only shows sessions scoped to the current working dir
# Issue draft: Plugin slash commands pass through as raw text in OpenCode Desktop v1.18.15 (regression) **Repo**: github.com/anomalyco/opencode **Type**: Bug / Regression **Component**: Desktop app command handling (plugin `command` hooks) ## Summary In **OpenCode Desktop v1.18.15**, plugin-registered slash commands (e.g. `/magic-compact 5`, `/magic-stats` f
### Description I was using opencode connected to DeepSeek V4 Flash Free on opencode zen, and most - *but not all*! - requests failed with an error, `Upstream request failed: [invalid_bearer_credential] Missing or invalid bearer credential` This happened with version 1.18.3. I upgraded to 1.18.18, and it kept happening. I then connected to DeepSeek V4 Flash
### Description As part of investigating what could've caused https://github.com/Opencode-DCP/opencode-dynamic-context-pruning/issues/407 I noticed that the tool was adding another system prompt, which caused for this model to break. Despite being fixed in the plugin it makes sense to at least have it as a known issue, so this is mostly for documentation pur
### Description The built-in `perplexity-agent` provider is currently failing on Opencode `1.17.4`. Any request to a Perplexity Agent model fails with: ``` AI_APICallError: invalid request body: json: unknown field "store" ``` I verified that the Perplexity Agent /v1/responses endpoint works if OpenAI-specific fields are omitted, but fails when fields such a
### Description Today, many OpenCode Go models interrupt their response mid-stream. I've observed this for DeepSeek-V4-Pro, Kimi-K3 and MiMo-V2-Pro. Other users report this issue for DeepSeek-V4-Flash. I am using the streaming API with my own client, so this is a server issue. I did not see any suspicious streaming packages in terms of `"data: ..."` `chat.co
## Summary OpenCode's cost tracking has multiple interrelated gaps that cause inaccurate cost display for users with multi-agent and multi-model workflows. This RFC proposes a unified architecture to address: 1. **Subagent cost aggregation** (#11027) - Parent sessions don't include child session costs 2. **Multi-model correctness** (#7387) - Costs recalculat
### Describe the bug When running `opencode upgrade` from behind a proxy/VPN, the command frequently fails with a 403 error due to GitHub API rate limiting for unauthenticated requests (60 requests/hour/IP). The command appears to make an anonymous request to `https://api.github.com/repos/anomalyco/opencode/releases/latest`. Even if the user has a valid `GIT
### Description The OpenCode Go docs use inconsistent model IDs for Kimi K2.7 Code. The model list and `/v1/models` endpoint use: ```text kimi-k2.7-code ``` But the endpoint table in the docs uses: ```text kimi-k2.7 ``` This makes it unclear which model ID should be used when configuring OpenCode Go manually or from external clients. The issue appears in mul
### Description I'm using a remote MCP server protected by OAuth 2.0 / OIDC. The initial authentication flow succeeds, and OpenCode stores both the access token and the refresh token. However, when the short-lived access token expires, OpenCode does not appear to perform a refresh-token grant. Instead, the MCP transitions to **"Requires Authentication"** and
### Description ### Describe the bug I purchased an OpenCode Go subscription, and the payment was successfully processed through Stripe. However, my workspace still reports: > Insufficient balance Because of this, I cannot use OpenCode Go even though the subscription has been paid. This appears to be a subscription provisioning or billing synchronization iss
### Description grok-4.5 has been returning errors since 2 Aug. ### Plugins NA ### OpenCode version v1.18.11 ### Steps to reproduce Call grok-4.5 on opencode go via openai chat completions api, response is always 500. ### Screenshot and/or share link NA ### Operating System Windows 11 ### Terminal Windows Terminal
### Description ### Description This appears to be a follow-up to #41990 The original issue occurs when a long conversation reaches the point where OpenCode starts summarizing/compacting the conversation and then reports that the usage limit has been reached. In my case, after the usage limit has eventually reset due to time passing, the affected conversatio
### Description ## Description The `read` tool incorrectly adds an extra space when handling file paths with Chinese characters. ### Plugins _No response_ ### OpenCode version 1.3.13 ### Steps to reproduce 1. File exists: `cline安装使用.md` , workpath:`/opt/tong/ws/git-repo/notes_superproject/temp/test` 2. send cmd: `cline安装使用.md 读取文件` 3. Call read tool 3. Retur
### Describe the bug Running `opencode upgrade` fails with an unexpected 403 HTTP status code when the GitHub API Rate Limit for unauthenticated requests from the client IP address is exhausted. This happens because the upgrade command queries `https://api.github.com/repos/anomalyco/opencode/releases/latest` using an unauthenticated request (without passing
### Description 在更新到 1.18.15 版本后，使用 deepseek v4 flash 官方 api 时出现了：通过 /compact 命令压缩上下文时只会输出 [Assistant tool call]: 一段shell命令 shell命令输出 类似这种结构，或者干脆就是只有一句“让我继续分析”完全没有对上下文的总结，导致第二轮对话AI忘记全部的上下文，需要重新阅读项目、重新了解现状。 目前使用过近5次compact全部是这种情况，猜测已经损坏而不是概率触发。 ### Plugins no ### OpenCode version 1.18.15 ### Steps to reproduce 1、接入deepseek官方api，思考模式改为high 2、进行多轮的问答 3、通过/compa
## Context CONTRIBUTING says UI changes need a design review before implementation, so raising this rather than sending a PR. The session picker shows footer hints for quick-switch and delete, but nothing for `app.toggle.session_directory_filter`. Even once the binding works in the picker (#42056) and is bindable, there is no indication in the picker that th
### Summary `deepseek-v4-pro` via OpenCode Go (`https://opencode.ai/zen/go/v1`, Responses API) returns HTTP 400 from Codex: ``` Error from provider (Console Go): Upstream request failed: [invalid_request_error] Failed to deserialize the JSON body into the target type: tools[8].function: missing field `name` ``` The same setup works with `deepseek-v4-flash`,
## Summary The current `type: "remote"` MCP configuration only supports **SSE (Server-Sent Events)** transport. However, the MCP protocol now includes **HTTP Streamable** transport (also called `http` or `streamable-http`), which is used by several major MCP servers including Sanity. ## Problem When configuring Sanity's hosted MCP server: ```json { "mcp": {
### Issue for this PR Closes #6231 ### Type of change - [ ] Bug fix - [x] New feature - [ ] Refactor / code improvement - [ ] Documentation ### What does this PR do? Adds a `local` provider to `CUSTOM_LOADERS` that auto-discovers models from any OpenAI-compatible `/v1/models` endpoint at startup. When configured with a `baseURL`, it fetches the model list, r
### Issue for this PR Closes #13715 Also addresses #7654. ### Type of change - [x] Bug fix - [ ] New feature - [ ] Refactor / code improvement - [ ] Documentation ### What does this PR do? Nested Task sessions can raise permission and question requests several levels below the root. The TUI, `opencode run`, and ACP previously matched only the session they di
### Description <img width="2504" height="693" alt="Image" src="https://github.com/user-attachments/assets/dd1d1dde-979c-4c85-86a3-29058db8de36" /> ### Plugins _No response_ ### OpenCode version _No response_ ### Steps to reproduce _No response_ ### Screenshot and/or share link _No response_ ### Operating System _No response_ ### Terminal _No response_
### Description <img width="1394" height="953" alt="Image" src="https://github.com/user-attachments/assets/0418c939-e71c-4239-864a-8d9a0c013612" /> <img width="1402" height="961" alt="Image" src="https://github.com/user-attachments/assets/9228a11b-d79c-4357-ad4e-3a5fe13c274d" /> When creating a new session (/new) and submitting the first prompt, the input is
### Feature hasn't been suggested before. - [x] I have verified this feature I am about to request hasn't been suggested before. ### Describe the enhancement you want to request **TL;DR**: Allow a model in a custom provider to reference/alias an official model definition from Models.dev by ID, so limit, modalities, reasoning, interleaved, cost, etc. are inhe
### 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 https://github.com/anomalyco/opencode/pull/30022 recently changed the oauth server to bind to 127.0.0.1 only. I use opencode inside a container, and was using `docker run -p 19876` to
## Summary In the OpenCode **desktop app**, file and folder paths rendered in assistant messages look like interactive chips/links, but they are not clickable. Markdown `file://` links are also not clickable. There is no way to open a referenced path in the built-in editor, system default app, or Finder/Explorer from the chat transcript. ## Environment - Ope
### Description opencode邀请失效了 <img width="1277" height="685" alt="Image" src="https://github.com/user-attachments/assets/4f511887-a5bc-4f90-8466-cd64b0392392" /> ### Plugins opencode cli ### OpenCode version 最新的 ### Steps to reproduce 1.邀请一个好友,好友使用你的订阅购买 2.然后等一段时间，本来可以换额度的，就被充值了 ### Screenshot and/or share link opencode邀请失效了 <img width="1300" height="699" al
## Summary Image attachments do not render correctly in the OpenCode TUI. The attachment is shown as an almost blank, dark thumbnail with only the `[Image 1]` label, so its visual content cannot be reviewed in the chat. ## Environment - opencode version: `v0.0.0-next-17151` - OS: Fedora Linux 44 (Workstation Edition), kernel `7.1.7-200.fc44.x86_64` - Termina
### Description Error: 429: {"type":"FreeUsageLimitError","message":"Error from provider (Console): Rate limit exceeded. Please try again later."} ?????????????? I have not used it today ### Plugins _No response_ ### OpenCode version _No response_ ### Steps to reproduce _No response_ ### Screenshot and/or share link _No response_ ### Operating System _No res
### Description Two separate things combine here, and I think they explain a large share of the "I paid for Go and it still says *Free usage exceeded, subscribe to Go*" reports (#36265, #42013, #42008, #42019, #40078, #34483, #39310, #31008, #29020, #27492 …). I could reproduce and measure both on macOS with Desktop 1.18.16. #### 1. `opencode` (Zen) and `ope
### Description When Tried to use free model can use perfectly but when a compactation or subagent triggers the opencode say "Free usage exceeded, subscribe to Go" but if i open another session it works but cant compact or use any subagents. been working fine problem appeared this afternoon. Also if i use another provider and compact then use the free openco
- [x] I have verified this feature I'm about to request hasn't been suggested before. # Describe the enhancement you want to request `opencode run` reads non-TTY stdin to EOF before starting a prompt. That is the right default for intentional input such as: ```sh cat context.txt | opencode run "review this" ``` The problem is that some supervisors give every
## Summary On opencode 1.18.15 (Arch package), the `github-copilot` provider never appears in the model picker. `opencode auth login -p github-copilot` works (auth succeeds), but `opencode models github-copilot` returns "Provider not found", and `/models` in the TUI shows no Copilot models. ## Root cause In `packages/opencode/src/plugin/github-copilot/copilo
### Description The Bug I Encountered: Its been exactly 24 hours since this is happening. When I open OpenCode or the ZEN API for Free Models, It is continuously saying: Free Usage Exceeded, Subscribe To Go, But its already been over 24 hours, The Daily Quota should be resetted. It works sometimes and fails 99% of the time. ### Plugins No Plugins ### OpenCod
### Issue for this PR Closes #30611 This replaces #30638, which was closed by automated cleanup. I rebased the change on the current `dev` branch, tightened the transport error handling, and limited retries for these failures. ### Type of change - [x] Bug fix - [ ] New feature - [ ] Refactor / code improvement - [ ] Documentation ### What does this PR do? Tr
## Summary - reorganize the settings navigation and split appearance and notifications into dedicated pages - add Figma-aligned Projects and Extensions views backed by real server configuration and MCP state - improve multi-server selection, local/default server ordering, and default-server menu behavior - add the supporting settings icons, layout styles, an
### Description I just installed OpenCode through the bash script but I can't use Ollama Local as I intended, because it's not showing up to connect. <img width="743" height="227" alt="Image" src="https://github.com/user-attachments/assets/1384b0dc-50ba-4b4a-a587-8c0ba382c258" /> ### Plugins _No response_ ### OpenCode version 1.3.3 ### Steps to reproduce _No
### Description I'm trying to use gemini models via a custom api but i get this error: `Invalid JSON payload received. Unknown name "ref" at 'tools[0].function_declarations[0].parameters.properties[0].value.items.properties[2].value.items': Cannot find field.` The provider config in opencode.json is this: ```json { "$schema": "https://opencode.ai/config.json
### 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 Currently, when using subagents in OpenCode, there is no visual indication in the TUI prompt footer about how many subagents are active or exist in total. Users have to manually navig
### 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 Problem Currently, /new (alias /clear) creates a brand new session with a new session ID. In Claude Code, /clear stays in the same session but clears the conversation context/history,
### 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 - [x] I have verified this feature does not already exist in the issue tracker. **Problem** The build script (`script/build.ts`) only supports `--single` to build for the current plat
### 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 Summary In subagent mode, there seem to be two scheduling / interaction issues: While subagents are running, the main session cannot accept new user input. When multiple subagents are