acp with Avante seems unresponsive on Darwin #3143

Closed
opened 2026-02-16 17:38:50 -05:00 by yindo · 11 comments
Owner

Originally created by @ReeSilva on GitHub (Nov 26, 2025).

Originally assigned to: @rekram1-node on GitHub.

Description

Hey everyone, good day, hope you're all doing good!

So, I've been using opencode acp for a while as my daily-driver ACP provider for Avante. After updating to version 1.0.110 (and then 1.0.114), opencode seems unresponsive when used. Actually, it goes to generating phase and then quickly Avante reports as succeeded, but nothing happens (you can see on the screenshot). Weirdly, this only happens in MacOS. When running in Nixos, it works well.

You can check the config that I use for Avante here: https://github.com/ReeSilva/Neve/blob/main/config/utils/avante.nix#L98

I think that, even though is Nix, it is simple enough to understand. But, if not, please let me know and I'll be happy to provide an equivalent.

Also, I've tried with two different model and provider (Codex from OpenAI and Gemini from Google), and both runs the same.

1.0.114

❯ nvim --version
NVIM v0.11.5
Build type: Release
LuaJIT 2.1.1741730670
Run "nvim -V1 -v" for more info
screenshot of avante with empty response from opencode's acp

OpenCode version

1.0.114

Steps to reproduce

  1. Configure opencode as an acp_provider in avante.nvim and make opencode your provider
require('avante').setup({
  input = {
    provider = "snacks",
    provider_opts = {
      title = "Avante Input",
      icon = " ",
      placeholder = "Enter your API key...",
    },
  },
  provider = "opencode",
  acp_providers = {
    opencode = {
      command = "opencode",
      args = ["acp"],
      env = {},
    },
  },
})
  1. Start a new session with :AvanteChatNew
  2. Send any message

Screenshot and/or share link

Image

Operating System

macOS 26.1 + nixos_unstable + nix-darwin master

Terminal

WezTerm commit(765d382)

Originally created by @ReeSilva on GitHub (Nov 26, 2025). Originally assigned to: @rekram1-node on GitHub. ### Description Hey everyone, good day, hope you're all doing good! So, I've been using `opencode acp` for a while as my daily-driver ACP provider for Avante. After updating to version 1.0.110 (and then 1.0.114), `opencode` seems unresponsive when used. Actually, it goes to `generating` phase and then quickly `Avante` reports as `succeeded`, but nothing happens (you can see on the screenshot). Weirdly, this only happens in MacOS. When running in Nixos, it works well. You can check the config that I use for Avante here: https://github.com/ReeSilva/Neve/blob/main/config/utils/avante.nix#L98 I think that, even though is Nix, it is simple enough to understand. But, if not, please let me know and I'll be happy to provide an equivalent. Also, I've tried with two different model and provider (Codex from OpenAI and Gemini from Google), and both runs the same. ```❯ opencode --version 1.0.114 ❯ nvim --version NVIM v0.11.5 Build type: Release LuaJIT 2.1.1741730670 Run "nvim -V1 -v" for more info ``` <img width="990" height="934" alt="screenshot of avante with empty response from opencode's acp" src="https://github.com/user-attachments/assets/33c49e81-d37e-4693-b220-3a81a17905ea" /> ### OpenCode version 1.0.114 ### Steps to reproduce 1. Configure `opencode` as an `acp_provider` in `avante.nvim` and make `opencode` your provider ```lua require('avante').setup({ input = { provider = "snacks", provider_opts = { title = "Avante Input", icon = " ", placeholder = "Enter your API key...", }, }, provider = "opencode", acp_providers = { opencode = { command = "opencode", args = ["acp"], env = {}, }, }, }) ``` 2. Start a new session with `:AvanteChatNew` 3. Send any message ### Screenshot and/or share link <img width="990" height="934" alt="Image" src="https://github.com/user-attachments/assets/d816f60f-73a4-42af-ade2-84008debbad7" /> ### Operating System macOS 26.1 + nixos_unstable + nix-darwin master ### Terminal WezTerm commit(765d382)
yindo added the opentuibugnix labels 2026-02-16 17:38:50 -05:00
yindo closed this issue 2026-02-16 17:38:50 -05:00
Author
Owner

@github-actions[bot] commented on GitHub (Nov 26, 2025):

This issue might be a duplicate of existing issues. Please check:

  • #3989: acp-command unhandled rejection - Similar ACP unhandled rejection errors, also reported on macOS with proxy settings
  • #2512: Opencode stuck in "generating" - Similar symptom of opencode getting stuck in generating state with no response
  • #4073: Hanging During Agent Tasks - macOS-specific hanging issues during agent execution, similar to the stuck generating state described here
  • #3868: Error: undefined is not an object (evaluating 'x3.data.providers') - Same error pattern mentioned in #3989, environment-specific (NixOS/Nix)
  • #4009: The response doesn't get show - Related issue where streamed content doesn't appear in the UI

Feel free to ignore if none of these address your specific case.

@github-actions[bot] commented on GitHub (Nov 26, 2025): This issue might be a duplicate of existing issues. Please check: - #3989: acp-command unhandled rejection - Similar ACP unhandled rejection errors, also reported on macOS with proxy settings - #2512: Opencode stuck in "generating" - Similar symptom of opencode getting stuck in generating state with no response - #4073: Hanging During Agent Tasks - macOS-specific hanging issues during agent execution, similar to the stuck generating state described here - #3868: Error: undefined is not an object (evaluating 'x3.data.providers') - Same error pattern mentioned in #3989, environment-specific (NixOS/Nix) - #4009: The response doesn't get show - Related issue where streamed content doesn't appear in the UI Feel free to ignore if none of these address your specific case.
Author
Owner

@ReeSilva commented on GitHub (Nov 27, 2025):

Still happens at 1.0.117

@ReeSilva commented on GitHub (Nov 27, 2025): Still happens at 1.0.117
Author
Owner

@rekram1-node commented on GitHub (Nov 27, 2025):

ill look into it

@rekram1-node commented on GitHub (Nov 27, 2025): ill look into it
Author
Owner

@ReeSilva commented on GitHub (Nov 27, 2025):

@rekram1-node just for adding more: I've switched from https://github.com/numtide/nix-ai-tools.git to opencode's own flake, now running under 1.0.119 and the error still exists, and with same pattern: work as expected in Nixos, "successfully fails" in Darwin.

@ReeSilva commented on GitHub (Nov 27, 2025): @rekram1-node just for adding more: I've switched from https://github.com/numtide/nix-ai-tools.git to opencode's own flake, now running under 1.0.119 and the error still exists, and with same pattern: work as expected in Nixos, "successfully fails" in Darwin.
Author
Owner

@rekram1-node commented on GitHub (Nov 28, 2025):

Are there any logs at all?

I know zed acp has ACP specific logs that help

You could also run opencode debug paths to see where opencode writes it's own logs

@rekram1-node commented on GitHub (Nov 28, 2025): Are there any logs at all? I know zed acp has ACP specific logs that help You could also run `opencode debug paths` to see where opencode writes it's own logs
Author
Owner

@ReeSilva commented on GitHub (Nov 28, 2025):

Are there any logs at all?

At least not at the interface. I'll find on avante.nvim if it is possible to enable logs.

You could also run opencode debug paths to see where opencode writes it's own logs

I'll do it 🫡

@ReeSilva commented on GitHub (Nov 28, 2025): > Are there any logs at all? At least not at the interface. I'll find on `avante.nvim` if it is possible to enable logs. > You could also run `opencode debug paths` to see where opencode writes it's own logs I'll do it 🫡
Author
Owner

@ReeSilva commented on GitHub (Nov 28, 2025):

So, logs were useful and turns out I was just being greedy and enabling too many MCP servers at the same time 😅 thx for the help, @rekram1-node

log for reference:

"stream":true,"stream_options":{"include_usage":true}},"statusCode":400,"responseHeaders":{"cf-placement":"remote-LAX","cf-ray":"REDACTED","connection":"keep-alive","content-type":"application/json","date":"REDACTED","server":"cloudflare","transfer-encoding":"chunked","vary":"accept-encoding"},"responseBody":"{\"message\":\"Maximum of 128 tools are supported, got 157\",\"type\":\"Bad Request\",\"code\":400}","isRetryable":false}} stream error
@ReeSilva commented on GitHub (Nov 28, 2025): So, logs were useful and turns out I was just being greedy and enabling too many MCP servers at the same time 😅 thx for the help, @rekram1-node log for reference: ``` "stream":true,"stream_options":{"include_usage":true}},"statusCode":400,"responseHeaders":{"cf-placement":"remote-LAX","cf-ray":"REDACTED","connection":"keep-alive","content-type":"application/json","date":"REDACTED","server":"cloudflare","transfer-encoding":"chunked","vary":"accept-encoding"},"responseBody":"{\"message\":\"Maximum of 128 tools are supported, got 157\",\"type\":\"Bad Request\",\"code\":400}","isRetryable":false}} stream error ```
Author
Owner

@rekram1-node commented on GitHub (Nov 28, 2025):

lol i need to improve the acp error handling they have a spec for it but im basically ignoring them

ill fix that when i get some time

@rekram1-node commented on GitHub (Nov 28, 2025): lol i need to improve the acp error handling they have a spec for it but im basically ignoring them ill fix that when i get some time
Author
Owner

@ReeSilva commented on GitHub (Nov 28, 2025):

btw, @rekram1-node food for thought (or not, maybe that's normal and I don't know enough to notice): funny the fact that it failed for opencode acp but works normally in tui (I use mcp-hub from ravitemer, so all my different ai interfaces have always the same set of mcp servers)

@ReeSilva commented on GitHub (Nov 28, 2025): btw, @rekram1-node food for thought (or not, maybe that's normal and I don't know enough to notice): funny the fact that it failed for `opencode acp` but works normally in `tui` (I use `mcp-hub` from `ravitemer`, so all my different ai interfaces have always the same set of mcp servers)
Author
Owner

@rekram1-node commented on GitHub (Nov 28, 2025):

I think its because the model it used in ACP didnt support it but the model u use in tui does. So it was just happenstance. EIther way need better error handling.

(That error isn't coming from us that comes from whatever provider got routed to)

@rekram1-node commented on GitHub (Nov 28, 2025): I think its because the model it used in ACP didnt support it but the model u use in tui does. So it was just happenstance. EIther way need better error handling. (That error isn't coming from us that comes from whatever provider got routed to)
Author
Owner

@ReeSilva commented on GitHub (Nov 28, 2025):

well, actually I always had the feeling that avante.nvim using acp would follow the model defined by the ACP itself. At least, when I change in the TUI, it changes on the acp as well.

@ReeSilva commented on GitHub (Nov 28, 2025): well, actually I always had the feeling that `avante.nvim` using `acp` would follow the model defined by the ACP itself. At least, when I change in the `TUI`, it changes on the acp as well.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#3143