fix(config): add azure-openai-responses to MODEL_APIS enum by Cypherm · Pull Request #52053 · openclaw/openclaw

Cypherm · 2026-03-22T03:42:49Z

Summary

Problem: models.providers.*.api config validation rejects "azure-openai-responses" because it's missing from the MODEL_APIS enum, even though pi-ai fully registers it as an API provider with correct Azure-specific auth handling (AzureOpenAI client, api-key header, api-version query param).

Why it matters: Models not in the built-in catalog (e.g., newer Azure models like gpt-5.4-mini) cannot be resolved through config. Setting api: "openai-responses" as a workaround routes to the plain OpenAI SDK client, causing wrong auth headers and HTTP 404 from Azure endpoints. This also breaks heartbeat and subagent sessions silently (0 input/0 output tokens).

What changed:

Added "azure-openai-responses" to the MODEL_APIS array in src/config/types.models.ts

What did NOT change:

Zod schema (zod-schema.core.ts) automatically picks up the change via z.enum(MODEL_APIS) — no manual update needed
ModelApi type automatically widens via (typeof MODEL_APIS)[number]
No runtime behavior change — pi-ai already registers and handles azure-openai-responses correctly
No changes to OPENAI_MODEL_APIS in transcript-policy.ts, OPENAI_RESPONSES_PROVIDERS in openai-stream-wrappers.ts, or any provider implementation
Config help text and labels are generic enough to cover the new value without changes

Change Type

Bug fix (non-breaking)

Scope

Gateway (config schema validation)

Linked Issue

Closes #51735

Security Impact

New permissions requested: none
Secrets handling changes: none
New network calls: none
New command/tool execution: none
Data access changes: none

Human Verification

I personally verified:

pnpm tsgo passes — ModelApi type correctly includes "azure-openai-responses"
pnpm test -- src/config/ — 106 test files, 843 tests pass
pnpm format:fix — no formatting issues
Verified z.enum(MODEL_APIS) in zod-schema.core.ts automatically includes the new value (line 182)

Evidence

Test Files  106 passed (106)
     Tests  843 passed (843)
  Duration  26.74s

git diff --stat upstream/main...HEAD: 1 file changed, 1 insertion(+)

What I Did NOT Verify

Not verified: end-to-end with a real Azure OpenAI endpoint and a model not in the built-in catalog
Not verified: heartbeat/subagent sessions with azure-openai-responses config (tested validation layer only)

Failure Recovery

If this breaks in production:

Detection: Config validation errors for azure-openai-responses would reappear
Rollback: Remove the one-line addition from MODEL_APIS
Blast radius: Only affects users configuring Azure OpenAI Responses API in openclaw.json — zero impact on users who don't use this provider

Generated with Claude Code

greptile-apps · 2026-03-22T03:44:53Z

Greptile Summary

This PR fixes a config validation gap by adding "azure-openai-responses" to the MODEL_APIS enum in src/config/types.models.ts. Without this entry, setting api: "azure-openai-responses" in openclaw.json was rejected by the Zod schema (z.enum(MODEL_APIS) at line 182 of zod-schema.core.ts), forcing users to use "openai-responses" as a workaround — which routes requests to the plain OpenAI SDK client and causes wrong auth headers and HTTP 404s against Azure endpoints.

The fix is minimal and correct: the ModelApi type ((typeof MODEL_APIS)[number]) and ModelApiSchema (z.enum(MODEL_APIS)) both update automatically, so no other schema changes are required.
pi-ai already fully registers and handles azure-openai-responses at the runtime level (visible in openai-stream-wrappers.ts OPENAI_RESPONSES_PROVIDERS), so this is purely a validation layer unlock.
The decision not to add "azure-openai-responses" to OPENAI_MODEL_APIS in transcript-policy.ts is correct: that set is only consulted as a fallback when no explicit provider is present, and azure-openai-responses is resolved at the plugin/provider layer by pi-ai when a provider is specified.
No other azure API variants (e.g., azure-openai-completions) appear in the codebase, confirming the PR is correctly scoped.

Confidence Score: 5/5

Safe to merge — minimal, correct, isolated change with clear validation evidence and zero blast radius for existing users.
Single-line enum addition with automatic downstream propagation through the type system and Zod schema. The underlying runtime support already exists in pi-ai. The PR description explains all intentionally-untouched files, and the reasoning is sound. 843 tests pass.
No files require special attention.

_{Last reviewed commit: "fix(config): add azu..."}

Cypherm · 2026-03-22T05:44:24Z

@jalehman — Config schema rejects azure-openai-responses for models.providers.*.api even though pi-ai registers it → added to MODEL_APIS enum. 1 file, +1 line. Closes #51735.

CI failures (channels, extension-fast (discord)) are pre-existing on main — same discord test regression in the latest main CI run.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6735fe390c

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-23T00:27:41Z

docs/.generated/config-baseline.jsonl

 {"recordType":"path","path":"models.providers","kind":"core","type":"object","required":false,"deprecated":false,"sensitive":false,"tags":["models"],"label":"Model Providers","help":"Provider map keyed by provider ID containing connection/auth settings and concrete model definitions. Use stable provider keys so references from agents and tooling remain portable across environments.","hasChildren":true}
 {"recordType":"path","path":"models.providers.*","kind":"core","type":"object","required":false,"deprecated":false,"sensitive":false,"tags":[],"hasChildren":true}
-{"recordType":"path","path":"models.providers.*.api","kind":"core","type":"string","required":false,"enumValues":["openai-completions","openai-responses","openai-codex-responses","anthropic-messages","google-generative-ai","github-copilot","bedrock-converse-stream","ollama"],"deprecated":false,"sensitive":false,"tags":["models"],"label":"Model Provider API Adapter","help":"Provider API adapter selection controlling request/response compatibility handling for model calls. Use the adapter that matches your upstream provider protocol to avoid feature mismatch.","hasChildren":false}
+{"recordType":"path","path":"models.providers.*.api","kind":"core","type":"string","required":false,"enumValues":["openai-completions","openai-responses","openai-codex-responses","azure-openai-responses","anthropic-messages","google-generative-ai","github-copilot","bedrock-converse-stream","ollama"],"deprecated":false,"sensitive":false,"tags":["models"],"label":"Model Provider API Adapter","help":"Provider API adapter selection controlling request/response compatibility handling for model calls. Use the adapter that matches your upstream provider protocol to avoid feature mismatch.","hasChildren":false}


Regenerate served config schema before updating enum baseline

This baseline line advertises azure-openai-responses, but the runtime schema served by config.schema is still sourced from GENERATED_BASE_CONFIG_SCHEMA (src/config/schema.ts:408-414), and that generated file still omits this enum value for models.providers.*.api and models.providers.*.models.*.api (src/config/schema.base.generated.ts:1029-1038 and 1133-1142). The commit therefore introduces a schema/baseline mismatch: tooling that relies on config.schema.lookup will not see the newly documented option, and pnpm config:docs:check can flag drift.

Useful? React with 👍 / 👎.

Addressed in the follow-up commit (chore: regenerate base config schema after rebase). The generated file src/config/schema.base.generated.ts now includes azure-openai-responses in both models.providers.*.api and models.providers.*.models.*.api enum arrays.

chatgpt-codex-connector · 2026-03-23T00:27:41Z

src/config/types.models.ts

  "openai-completions",
  "openai-responses",
  "openai-codex-responses",
+  "azure-openai-responses",


Handle azure-openai-responses in Responses API guard paths

Allowing "azure-openai-responses" in MODEL_APIS makes this value reachable from user config, but multiple runtime guards still only treat "openai-responses"/"openai-codex-responses" as Responses APIs (for example sanitizeSessionHistory in src/agents/pi-embedded-runner/google.ts:566-583, stream wrapping in src/agents/pi-embedded-runner/run/attempt.ts:2320-2341, and OPENAI_RESPONSES_APIS in src/agents/pi-embedded-runner/openai-stream-wrappers.ts:11-159). Those guards apply transcript/payload downgrades required to avoid Responses API rejections (see src/agents/pi-embedded-helpers/openai.ts:85-209), so configs using the new enum can still fail on tool/reasoning replay flows.

Useful? React with 👍 / 👎.

Acknowledged. The Responses API guard paths (sanitizeSessionHistory, stream wrapping, etc.) are a pre-existing gap — they don't handle azure-openai-responses yet, but they also don't handle it on main today. This PR's scope is strictly the config validation layer (allowing the value through schema). The runtime guard paths are a separate concern tracked in the original issue discussion. Happy to address in a follow-up if maintainers want it bundled.

Cypherm · 2026-03-23T00:37:15Z

CI failure in check (tsgo TS1117) is unrelated to this PR:

Failed: tsgo reports TS1117: An object literal cannot have multiple properties with the same name in src/cli/acp-cli.option-collisions.test.ts, cron-cli.test.ts, devices-cli.test.ts, etc.
Our changes: exclusively in src/config/types.models.ts (+1 line) and src/config/schema.base.generated.ts (regen)
Same errors on main: confirmed identical TS1117 failures on main's latest check run
check-additional fails because it gates on check passing

Happy to rebase once upstream fixes the tsgo errors.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f33cca1a65

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-23T05:00:15Z

src/config/types.models.ts

  "openai-completions",
  "openai-responses",
  "openai-codex-responses",
+  "azure-openai-responses",


Update Responses guard lists for azure-openai-responses

Adding "azure-openai-responses" to MODEL_APIS makes that value valid in user config, but the runtime still gates critical Responses handling on "openai-responses"/"openai-codex-responses" only (for example OPENAI_RESPONSES_APIS in src/agents/pi-embedded-runner/openai-stream-wrappers.ts:11, isOpenAIResponsesApi in src/agents/pi-embedded-runner/google.ts:566-567, and the downgrade wrapper branch in src/agents/pi-embedded-runner/run/attempt.ts:2320-2323). That means configs newly accepted by this commit can skip store/context and transcript downgrade logic used to prevent Responses API rejections, so Azure Responses runs can regress on tool/replay flows. Fresh evidence vs earlier review: this commit also updates src/config/schema.base.generated.ts to include the enum, so the value is now reachable through validated config paths.

Useful? React with 👍 / 👎.

Same concern as the prior reply (see comment above). This PR's scope is strictly the config validation layer — allowing azure-openai-responses through the schema. The runtime guard paths (OPENAI_RESPONSES_APIS, etc.) are a pre-existing gap that exists on main today and are tracked separately. Happy to address in a follow-up if maintainers want it bundled.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4bbf189f86

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-23T06:55:11Z

src/config/types.models.ts

  "openai-completions",
  "openai-responses",
  "openai-codex-responses",
+  "azure-openai-responses",


Extend Responses guard paths for azure-openai-responses

Adding azure-openai-responses to MODEL_APIS makes models.providers.*.api/models.providers.*.models.*.api configs validate, but the runtime Responses safety logic still only matches openai-responses/openai-codex-responses (src/agents/pi-embedded-runner/openai-stream-wrappers.ts:11-159, src/agents/pi-embedded-runner/google.ts:566-583, src/agents/pi-embedded-runner/run/attempt.ts:2318-2338). With api: "azure-openai-responses", those downgrade/store-guard branches are skipped, so Azure Responses requests can miss the replay/payload sanitization that prevents multi-turn/tool-call rejections. Fresh evidence: the existing Azure regression coverage only asserts the store-force behavior for api: "openai-responses" (src/agents/pi-embedded-runner-extraparams.test.ts:1729-1741), which means this newly valid enum path is currently unprotected.

Useful? React with 👍 / 👎.

Same concern as the prior two replies (see above). This PR's scope is the config validation layer only. The runtime Responses guard paths are a pre-existing gap on main, not introduced by this PR.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 65cc7078ca

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-23T09:29:53Z

src/config/types.models.ts

  "openai-completions",
  "openai-responses",
  "openai-codex-responses",
+  "azure-openai-responses",


Update Responses guards for azure-openai-responses API

Adding "azure-openai-responses" to MODEL_APIS makes models.providers.*.api/models.providers.*.models.*.api accept this value, and applyConfiguredProviderOverrides propagates provider api to the runtime model (src/agents/pi-embedded-runner/model.ts:134), so this commit introduces a newly reachable runtime state where model.api === "azure-openai-responses". The OpenAI Responses safety wrappers still gate on openai-responses only (OPENAI_RESPONSES_APIS in src/agents/pi-embedded-runner/openai-stream-wrappers.ts), which means the store=true and context-management payload patching used to prevent Azure Responses multi-turn failures is skipped for this new config path.

Useful? React with 👍 / 👎.

Same concern as prior replies — see above.

Cypherm · 2026-03-24T00:25:02Z

CI triage update: 3 remaining failures are upstream — same failures present on main.

Failing (upstream):

check (tsgo) — shared lint/type-check, not our files
check-additional — gates on check
checks (node, channels) — channel tests, unrelated to config schema

Our area all passes:

✅ build-smoke — builds successfully
✅ checks (node, test, 1/2) — main test suite shard 1
✅ checks (node, test, 2/2) — main test suite shard 2
✅ checks (node, extensions) — extension tests
✅ checks (node, contracts) — contract tests
✅ checks-windows — 5/5 shards pass
✅ check-docs — docs scope clean

Our change: +1 line in src/config/types.models.ts + schema regen. All test suites covering config pass.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 509b6eb77d

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-30T00:43:13Z

src/config/types.models.ts

  "openai-completions",
  "openai-responses",
  "openai-codex-responses",
+  "azure-openai-responses",


Wire new API enum through model transport normalization

Adding "azure-openai-responses" to MODEL_APIS makes this config value valid, but the runtime transport whitelist still drops it (normalizeResolvedTransportApi in src/agents/pi-embedded-runner/model.ts:92-105). As a result, provider-level configs like models.providers.<id>.api: "azure-openai-responses" are silently coerced back to openai-responses in fallback/override paths (model.ts:322-334, 365-375, 528-543), so requests can route through the wrong OpenAI transport instead of Azure Responses. This means the new accepted enum value does not actually produce the intended runtime behavior for common provider-level configuration.

Useful? React with 👍 / 👎.

Same concern as prior replies — see above.

Cypherm · 2026-03-30T03:32:52Z

CI failures are unrelated to this PR:

Our changes: src/config/types.models.ts (MODEL_APIS enum), src/config/schema.base.generated.ts, and regenerated doc baselines. All config-scoped.

Checks covering our area — all pass:

check ✅ (full type check)
check-additional ✅
check-docs ✅
checks-fast-contracts-protocol ✅
checks-node-test-1 ✅
security-fast ✅

Failing checks — all in unrelated areas:

checks-fast-extensions: resetTelegramThreadBindingsForTests is not a function in extensions/matrix/src/registry-backed.contract.test.ts — upstream telegram plugin refactor
checks-node-test-2: install.test.ts daemon token mock assertion — upstream CLI changes
build-smoke: CLI startup memory check — infrastructure
checks-node-channels-*: test-parallel infrastructure failures in discord/message-utils — upstream
checks-windows-*: same infrastructure failures on Windows

Main is currently running 5 CI jobs simultaneously — active refactoring by maintainers is the source of these failures. Happy to rebase once main stabilizes.

Azure OpenAI Responses API is registered as a provider in pi-ai but was missing from the config schema enum, causing validation failure when users set models.providers.*.api to 'azure-openai-responses'. Models not in the built-in catalog could not be resolved through config. Closes openclaw#51735

Cypherm · 2026-03-31T03:57:48Z

Closing — the core change (adding azure-openai-responses to MODEL_APIS) was already merged upstream via #50851 by @kunalk16. After rebasing, this branch has no net diff vs main. Thanks!

openclaw-barnacle bot added the size: XS label Mar 22, 2026

openclaw-barnacle bot added the docs Improvements or additions to documentation label Mar 22, 2026

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch from 7d7ac8c to 6735fe3 Compare March 23, 2026 00:21

chatgpt-codex-connector bot reviewed Mar 23, 2026

View reviewed changes

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch from 47f8f2e to f33cca1 Compare March 23, 2026 04:55

chatgpt-codex-connector bot reviewed Mar 23, 2026

View reviewed changes

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch from f33cca1 to 4bbf189 Compare March 23, 2026 06:47

chatgpt-codex-connector bot reviewed Mar 23, 2026

View reviewed changes

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch 2 times, most recently from db1d427 to 65cc707 Compare March 23, 2026 09:23

chatgpt-codex-connector bot reviewed Mar 23, 2026

View reviewed changes

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch 2 times, most recently from 2f1c3e9 to cd4634a Compare March 23, 2026 15:13

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch from cd4634a to 80df3a0 Compare March 24, 2026 08:12

BradGroux mentioned this pull request Mar 26, 2026

WORKING: All Microsoft Issues and PRs #49126

Draft

Copilot AI mentioned this pull request Mar 27, 2026

List Azure and Windows open PRs available for review johnsonshi/openclaw#1

Closed

17 tasks

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch from 1da4263 to 509b6eb Compare March 30, 2026 00:38

chatgpt-codex-connector bot reviewed Mar 30, 2026

View reviewed changes

Cypherm added 7 commits March 31, 2026 11:49

chore: regenerate config docs baseline after MODEL_APIS change

9769ab6

chore: regenerate base config schema after rebase

4d5b70f

fix: regenerate plugin-sdk-api baseline after rebase

088ab9c

chore: regenerate config docs baseline after rebase

9d92ada

chore: regenerate plugin-sdk-api baseline after rebase

f226b65

chore: regenerate baselines after rebase onto upstream/main

cefa21c

Cypherm force-pushed the fix/config-azure-openai-responses-api-enum branch from 509b6eb to cefa21c Compare March 31, 2026 03:53

Cypherm closed this Mar 31, 2026

altaywtf mentioned this pull request Apr 5, 2026

Config schema missing 'azure-openai-responses' in models.providers.*.api enum #51735

Closed

Uh oh!

Conversation

Cypherm commented Mar 22, 2026

Summary

Change Type

Scope

Linked Issue

Security Impact

Human Verification

Evidence

What I Did NOT Verify

Failure Recovery

Uh oh!

greptile-apps bot commented Mar 22, 2026

Greptile Summary

Confidence Score: 5/5

Uh oh!

Cypherm commented Mar 22, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm commented Mar 23, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm commented Mar 24, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm Mar 31, 2026

Choose a reason for hiding this comment

Uh oh!

Cypherm commented Mar 30, 2026