Is there a way to disable streaming mode?

Can OpenCode work without Streaming mode? The proxy provider I'm using doesn't support streaming at the moment:

AI_APICallError: Streaming is not supported in Credal OpenAI Proxy

+1, because you now need to be verified organization to use steaming with gpt5, so being able to turn off streaming would be great.

Facing the same issue

+1

Is this something an outside contribution could solve or does it need a lot of work?

im also interested in testing non-streaming mode for opencode, that may save some network, making debugging of problems in the protocol easier too.

+1

feat(opencode): respect provider/model streaming:false to disable res…

FYI: For those of you having problem using Gemma 4 31B on vLLM with OpenCode. Updating vLLM to 0.24.0 and using the vLLM example template solved all of our issues.

FYI: For those of you having problem using Gemma 4 31B on vLLM with OpenCode. Updating vLLM to 0.24.0 and using the vLLM example template solved all of our issues.

Any clue how to make DeepSeek v4 Flash 0731 work with streaming? It's the same issue there: corrupted tool calls. Both on vLLM and sglang (used latest dev nightly builds).

FYI: For those of you having problem using Gemma 4 31B on vLLM with OpenCode. Updating vLLM to 0.24.0 and using the vLLM example template solved all of our issues.

Any clue how to make DeepSeek v4 Flash 0731 work with streaming? It's the same issue there: corrupted tool calls. Both on vLLM and sglang (used latest dev nightly builds).

Haven't been able to test, as we're not allowed to deploy Chinese models at work, unfortunately...

@izogfif

@izogfif Use a litellm proxy

model_list:
    -  model_name: deepseek-ai/DeepSeek-V4-Flash
       litellm_params:
            model: hosted_vllm/deepseek-ai/DeepSeek-V4-Flash-0731
            api_base: http://localhost:8000/v1
            api_key: dummy

@junwha

@junwha

Use a litellm proxy

Does it generate fake / empty streaming tokens in response? Opencode has some kind of built-in timeout (around 300 seconds, I think). If it doesn't receive at least one streaming response token within that limit, it will consider API endpoint timed out and retry the request. And I often encounter situations when it takes more than 300 seconds to generate the full non-streaming response.
Here is what Google AI told me but I'm not sure if it's true:

While LiteLLM does have an internal "fake streaming" mechanism (MockResponsesAPIStreamingIterator) that translates a stream=true request into a non-streaming upstream call, it waits for the complete upstream non-streaming response to finish before it splits that response up into chunks and sends them back to your consumer. It does not proactively emit periodic "keep-alive" or empty filler tokens to the client while the upstream endpoint is still thinking.

To prevent opencode from timing out while waiting for a slow non-streaming backend, you will need to intercept the request and handle the keeping-alive yourself.

Sources :

[Feature]: Support enabling fake streaming generally for all providers, or at least for OpenAI BerriAI/litellm#30208

[Bug]: Responses API streaming falls back to fake stream for custom models, silently dropping function_call events BerriAI/litellm#21090

[Feature]: Force enable fake streaming per model BerriAI/litellm#5416

Reminder: https://github.com/crashr/llama-stream could be a workaround.

+1

@crashr

Thanks for the request, and to everyone who added their use cases. The ask is an option to have OpenCode send non-streaming requests (stream: false) to a provider or model. That would cover proxies and backends that can't stream, like the Credal OpenAI Proxy, and backends whose streamed output is broken.

Right now OpenCode always streams model responses, including in v2, and there is no config setting to turn that off. Maintainers will consider adding one. Until then, the workaround people here have used is a small local proxy that sends stream: false upstream and re-wraps the response as SSE (for example the one linked above by @crashr).

feat: add streaming option to provider config to support non-streaming backends #14719 and feat(opencode): support non-streaming provider mode + fallback tests #15828: earlier community PRs that tried to add a provider-level streaming: false option (both closed without merging)

Thanks for the suggestion. This isn't something we plan to build, so closing to keep the tracker focused.

40 upvotes, dozens of +1, PRs that fix this....
TBH, that is the way to move people away

Navigation Menu

Repository navigation

Is there a way to disable streaming mode? #785

14 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Is there a way to disable streaming mode? #785

Description

Activity

McAvity commented on Aug 8, 2025

harkairt commented on Aug 10, 2025

archodev commented on Aug 17, 2025

jordanbaindev commented on Sep 8, 2025

LeonMusCoden commented on Sep 15, 2025

JeongJuhyeon commented on Sep 25, 2025

trufae commented on Oct 16, 2025

HansuHessu commented on Oct 25, 2025

dayeguilaiye commented on Oct 27, 2025

jayteaftw commented on Nov 7, 2025

eleqtrizit commented on Jan 17, 2026

csawtelle commented on Jan 23, 2026

smetay commented on Feb 2, 2026

RawthiL commented on Feb 3, 2026

14 remaining items

sebdanielsson commented on Jul 3, 2026

izogfif commented on Aug 9, 2026

sebdanielsson commented on Aug 10, 2026

junwha commented on Aug 21, 2026

izogfif commented on Aug 28, 2026

crashr commented on Aug 28, 2026

StroepWafel commented on Sep 24, 2026

opencodejr commented on Oct 29, 2026

neriousy commented on Nov 12, 2026

poelzi commented on Nov 13, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions