OpenAI responses streaming #5837

katzdave · 2025-11-21T20:38:15Z

This PR implements support for OpenAI's Responses API with streaming

Implemented full support for OpenAI's Responses API format with streaming capabilities
Added new openai_responses.rs format handler for the responses endpoint
Updated OpenAI provider to handle both chat completions and responses formats
Improved token counting and usage tracking for the new format

Co-authored-by: Copilot <[email protected]>

'

…nses-streaming * 'main' of github.com:block/goose: fix: adjust strange spacing in agent.rs (#5877)

@goose

…nses-streaming * 'main' of github.com:block/goose: Feat/automatic update installation (#5345) fix: Added "Merged consecutive assistant messages" to the acceptable issues for moim injection check (#5933) fix: anthropic provider model fetching (#5932) [MCP-UI] add CSP for images to proxy HTML (#5931) fix: correct typo in blog post (AIMDOEL -> AIMODEL) (#5902) feat: @goose in terminal (native terminal support) (#5887) docs: adding AI-friendly features (#5918) Blog/advent of ai announcement (#5917) Extension selector behind ALPHA flag (#5892) blog: typo fixes (#5896) blog: fixing img url (#5895) blog: MCPs for Developers (#5884) docs: Extension Manager MCP (#5883) Update cleanup marker logic for Fedora users. (#5868) Improve AWS credential loading and configuration handling in BedrockProvider (#5699)

…nses-streaming * 'main' of github.com:block/goose: docs: promote planning guide (#5934) fix: use a lock to ensure only need to run tunnel just in case multiple go… (#5885) Feat: Added custom headers and toggle keyring CLI options (#5017)

katzdave · 2025-12-02T00:52:58Z

Still figuring out how to get usage data for the streaming, but otherwise this is working pretty well. Interestingly usage comes through when there's a tool call.

katzdave · 2025-12-02T23:52:06Z

@michaelneale This seems to be working pretty well for me. Want to take it for test drive?

…nses-streaming * 'main' of github.com:block/goose: blog: MCP Sampling (#5987) Update Anthropic and Google Gemini models to latest API versions (#5980) docs: chat recall tutorial (#5975) fix: `final assistant content cannot end with trailing whitespace` error from Anthropic (#5967) 5527 multiple file popups (#5905) Groq configure fix (#5833) added sidebar contextual information for firefox (#5433) docs: council of mine MCP (#5979) docs: nano banana extension (#5977) fix: remove prompt change, read model from config (#5976) Enable recipe deeplink parameters for pre-population (#5757) chore: upgrade npm packages (#5951) feat: ActionRequired (#5897) feat(acp): support loading sessions in acp (#5942) docs: add videos to multi-model page (#5938)

…nses-streaming * 'main' of github.com:block/goose: Onboarding detect provider from api key (#5955) Fix PATH on Windows for extensions (#6000) recipe(datahub): Add a recipe for searching & understanding data in DataHub (#5859) fix params not being substituted in activities (#5992)

…nses-streaming * 'main' of github.com:block/goose: blog: How to stop your ai agent from making unwanted code changes (#6055) Fix typo in prompt (#6007) Docs for new terminal integration feature (#6042) fix: linux goosed crashing libssl error (#6051) chore(release): release version 1.16.0 (minor) (#5985) fix: metrics on posthog (#6024) gov: new LF Projects LLC section (#6027) Cleanup: Remove Recipe Key Flow (#6015) chore(deps): bump mdast-util-to-hast from 13.2.0 to 13.2.1 in /documentation (#5963) remove problematic corrupted woff font (#6006) Added search bar / filtering for recipes (#6019) Hide recipe icon in empty chat (#6022) docs: provider and model config (#6008) Show modal selector after configuring a provider (#6005) docs: additional mcp sampling resources (#6020) Flutter PR Code Review (#6011) feat(mcp): elicitation support (#5965)

Copilot

Pull request overview

This PR implements support for OpenAI's new Responses API format with streaming capabilities, enabling Goose to work with newer OpenAI models (gpt-5-codex and gpt-5.1-codex) that use a different API format than the traditional Chat Completions endpoint.

Key changes:

Added a new format handler for the OpenAI Responses API with full streaming support
Updated the OpenAI provider to route requests based on model name, using the Responses API for gpt-5-codex models
Implemented proper token counting and usage tracking for the new format

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
crates/goose/src/providers/openai.rs	Added model detection logic to route gpt-5-codex models to the Responses API, with separate handling for both complete and stream methods
crates/goose/src/providers/formats/openai_responses.rs	New format handler implementing full Responses API support including request creation, response parsing, streaming, and usage tracking
crates/goose/src/providers/formats/mod.rs	Registered the new openai_responses module

michaelneale

nice one @katzdave tested it with gpt-5.2 and codex 5.1 and worked nicely! awesome

cbruyndoncx · 2025-12-12T10:41:43Z

@michaelneale Hi guys , I am doing a project where i need deep research, what is the outlook for this landing in the next release ?

DOsinga

let's remove some of the obvious comments, but other than that, great

one concern I have is how we handle base_path. for these models we just ignore it. but if the user overrides it through a custom provider, we probably shouldn't. not sure what the best way of handling this is, tbh. in some ideal world, we would never have set the base_path to v1/ instead of the whole path and then we could append

so one way could be to detect whether the base_path has been overwritten instead of giving it a default and then when we need it, see if ends on /completions/reply or what have you and replace that with what we want and if it doesn't, just use it as is?

DOsinga · 2025-12-12T13:25:47Z

crates/goose/src/providers/formats/openai_responses.rs

+                &response_str
+            };
+
+            // Skip [DONE] marker


I can see that

DOsinga · 2025-12-12T13:26:27Z

crates/goose/src/providers/formats/openai_responses.rs

+                }
+
+                ResponsesStreamEvent::ResponseCompleted { response, .. } => {
+                    // Always set final usage (use default if not provided)


DOsinga · 2025-12-12T13:30:24Z

crates/goose/src/providers/openai.rs

    }

+    fn uses_responses_api(model_name: &str) -> bool {
+        model_name.starts_with("gpt-5-codex") || model_name.starts_with("gpt-5.1-codex")


after your model normalization this might work, but right now, does this cover like databricks? also, people might change the base_path to v1/responses for models that support both

I don't actually see the codex models inside of databricks. I gave goose-gpt-5.2 a try and it failed to configure very cryptically with:

Request failed: Bad request (400): {"external_model_provider":"databricks-model-serving","external_model_error":{"error_code":"BAD_REQUEST","message":"BAD_REQUEST: Invalid role in the chat message. The role must be one of: "system", "user", "assistant", "tool", "function"."}}

Saw there was another thread about that.

Good callout on the base path. We probably need to parse that + also be aware about which models support both.

I did a quick test of the openrouter endpoint I'm using for canonical models and seems like you can infer which models support completions/responses based on the supported_parameters part of the payload.

So I think we follow up with adding something to canonical model around supported_formats.

For the openai provider, if the model supports both, trust the user on the base path. If the model only supports responses, continue to modify the base path suffix for them.

…oose into dkatz/openai-responses-streaming * 'dkatz/openai-responses-streaming' of github.com:block/goose: Rm stray leave errors in the context blog: How to stop your ai agent from making unwanted code changes (#6055) Fix typo in prompt (#6007) Docs for new terminal integration feature (#6042) fix: linux goosed crashing libssl error (#6051) chore(release): release version 1.16.0 (minor) (#5985) fix: metrics on posthog (#6024) gov: new LF Projects LLC section (#6027) Cleanup: Remove Recipe Key Flow (#6015) chore(deps): bump mdast-util-to-hast from 13.2.0 to 13.2.1 in /documentation (#5963) remove problematic corrupted woff font (#6006) Added search bar / filtering for recipes (#6019) Hide recipe icon in empty chat (#6022) docs: provider and model config (#6008) Show modal selector after configuring a provider (#6005) docs: additional mcp sampling resources (#6020) Flutter PR Code Review (#6011) feat(mcp): elicitation support (#5965)

…nses-streaming * 'main' of github.com:block/goose: Fix community page mobile responsiveness and horizontal overflow (#6082) Tool reply meta (#6074) chore: avoid accidentally using native tls again (#6086) Update vars to be capitalised to be in line with other variables in config file (#6085) docs: restructure recipe reference (#5972) docs: configure custom providers (#6044) docs: Community All-Stars Spotlight November 2025, CodeTV Hackathon edition (#6070) fix: include file attachments in queued messages (#5961) fix(ui): prevent incorrect provider type suffix in update dialog #5908 (#5909) docs: mcp elicitation (#6060)

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

* main: fix: Make datetime info message more explicit to prevent LLM confusion about current year (#6101) refactor: unify subagent and subrecipe tools into single tool (#5893) goose repo is too big for the issue solver workflow worker (#6099) fix: use system not developer role in db (#6098) Add /goose issue solver github workflow (#6068) OpenAI responses streaming (#5837) Canonical models for Providers (#5694) feat: add Inception provider for Mercury models (#6029) fix old sessions with tool results not loading (#6094) Fix community page mobile responsiveness and horizontal overflow (#6082)

* 'main' of github.com:block/goose: (22 commits) Disallow subagents with no extensions (#5825) chore(deps): bump js-yaml in /documentation (#6093) feat: external goosed server (#5978) fix: Make datetime info message more explicit to prevent LLM confusion about current year (#6101) refactor: unify subagent and subrecipe tools into single tool (#5893) goose repo is too big for the issue solver workflow worker (#6099) fix: use system not developer role in db (#6098) Add /goose issue solver github workflow (#6068) OpenAI responses streaming (#5837) Canonical models for Providers (#5694) feat: add Inception provider for Mercury models (#6029) fix old sessions with tool results not loading (#6094) Fix community page mobile responsiveness and horizontal overflow (#6082) Tool reply meta (#6074) chore: avoid accidentally using native tls again (#6086) Update vars to be capitalised to be in line with other variables in config file (#6085) docs: restructure recipe reference (#5972) docs: configure custom providers (#6044) docs: Community All-Stars Spotlight November 2025, CodeTV Hackathon edition (#6070) fix: include file attachments in queued messages (#5961) ... # Conflicts: # crates/goose-server/src/routes/agent.rs # crates/goose/src/agents/extension_manager.rs # ui/desktop/src/api/sdk.gen.ts

Co-authored-by: Michael Neale <[email protected]> Co-authored-by: Copilot <[email protected]>

…sions * 'main' of github.com:block/goose: (22 commits) Disallow subagents with no extensions (#5825) chore(deps): bump js-yaml in /documentation (#6093) feat: external goosed server (#5978) fix: Make datetime info message more explicit to prevent LLM confusion about current year (#6101) refactor: unify subagent and subrecipe tools into single tool (#5893) goose repo is too big for the issue solver workflow worker (#6099) fix: use system not developer role in db (#6098) Add /goose issue solver github workflow (#6068) OpenAI responses streaming (#5837) Canonical models for Providers (#5694) feat: add Inception provider for Mercury models (#6029) fix old sessions with tool results not loading (#6094) Fix community page mobile responsiveness and horizontal overflow (#6082) Tool reply meta (#6074) chore: avoid accidentally using native tls again (#6086) Update vars to be capitalised to be in line with other variables in config file (#6085) docs: restructure recipe reference (#5972) docs: configure custom providers (#6044) docs: Community All-Stars Spotlight November 2025, CodeTV Hackathon edition (#6070) fix: include file attachments in queued messages (#5961) ... # Conflicts: # crates/goose-server/src/routes/agent.rs # crates/goose/src/agents/extension_manager.rs # ui/desktop/src/api/sdk.gen.ts

michaelneale and others added 7 commits November 18, 2025 12:38

first cut of responses

7c018f2

remove slop

03d8d81

cleanup

4d9800c

Update crates/goose/src/providers/openai.rs

2981610

Co-authored-by: Copilot <[email protected]>

working streaming

733dfdd

'

v2

992621b

rm completions pack

57d8093

katzdave mentioned this pull request Nov 21, 2025

support for openai responses api #5783

Closed

katzdave changed the base branch from micn/openai-responses to main November 24, 2025 20:24

katzdave added 7 commits November 25, 2025 17:33

merge

6be9b45

Merge branch 'main' of github.com:block/goose into dkatz/openai-respo…

98a5f2e

…nses-streaming * 'main' of github.com:block/goose: fix: adjust strange spacing in agent.rs (#5877)

fix delta

2222251

fix main client message issue

eed457e

more improvements, token counting still broken

4c713eb

katzdave added 8 commits December 1, 2025 20:27

token counter works! But not first message

3c0022d

rm noise

3f668a5

fmt

45defdd

structured tool calls working

b66ee73

cleanup warnings and last response id

89c1f2a

lint

9980f80

hopefully fix lint

e9cdab8

fmt

3a2cbca

katzdave added 4 commits December 3, 2025 12:34

separate the format

f8c572a

provider usage diff

e8bdfe7

katzdave added 4 commits December 5, 2025 17:14

rm more comments

3286027

more delete

907900c

leave errors in the context

e1af529

Rm stray

d5df556

katzdave marked this pull request as ready for review December 10, 2025 23:35

Copilot AI review requested due to automatic review settings December 10, 2025 23:35

Copilot started reviewing on behalf of katzdave December 10, 2025 23:36 View session

Copilot AI reviewed Dec 10, 2025

View reviewed changes

katzdave requested review from michaelneale and tlongwell-block December 11, 2025 00:29

michaelneale self-assigned this Dec 12, 2025

michaelneale approved these changes Dec 12, 2025

View reviewed changes

DOsinga approved these changes Dec 12, 2025

View reviewed changes

katzdave added 3 commits December 12, 2025 11:55

rm comments

9e6300a

Copilot AI review requested due to automatic review settings December 12, 2025 17:36

Copilot started reviewing on behalf of katzdave December 12, 2025 17:36 View session

build error

f3c24fc

Copilot AI reviewed Dec 12, 2025

View reviewed changes

katzdave merged commit 239f4ec into main Dec 12, 2025
17 checks passed

katzdave deleted the dkatz/openai-responses-streaming branch December 12, 2025 22:25

github-actions bot mentioned this pull request Dec 16, 2025

chore(release): release version 1.17.0 (minor) #6131

Closed

fbalicchia pushed a commit to fbalicchia/goose that referenced this pull request Dec 16, 2025

OpenAI responses streaming (block#5837)

d540bc3

Co-authored-by: Michael Neale <[email protected]> Co-authored-by: Copilot <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

OpenAI responses streaming #5837

OpenAI responses streaming #5837

Uh oh!

katzdave commented Nov 21, 2025 •

edited

Loading

Uh oh!

katzdave commented Dec 2, 2025

Uh oh!

katzdave commented Dec 2, 2025

Uh oh!

Copilot AI left a comment

Uh oh!

michaelneale left a comment

Uh oh!

cbruyndoncx commented Dec 12, 2025

Uh oh!

DOsinga left a comment

Uh oh!

DOsinga Dec 12, 2025

Uh oh!

DOsinga Dec 12, 2025

Uh oh!

DOsinga Dec 12, 2025

Uh oh!

katzdave Dec 12, 2025 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

OpenAI responses streaming #5837

OpenAI responses streaming #5837

Uh oh!

Conversation

katzdave commented Nov 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

katzdave commented Dec 2, 2025

Uh oh!

katzdave commented Dec 2, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

michaelneale left a comment

Choose a reason for hiding this comment

Uh oh!

cbruyndoncx commented Dec 12, 2025

Uh oh!

DOsinga left a comment

Choose a reason for hiding this comment

Uh oh!

DOsinga Dec 12, 2025

Choose a reason for hiding this comment

Uh oh!

DOsinga Dec 12, 2025

Choose a reason for hiding this comment

Uh oh!

DOsinga Dec 12, 2025

Choose a reason for hiding this comment

Uh oh!

katzdave Dec 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

katzdave commented Nov 21, 2025 •

edited

Loading

katzdave Dec 12, 2025 •

edited

Loading