fix(aionrs): default missing model protocol to openai by jiahe0510 · Pull Request #577 · iOfficeAI/AionCore

jiahe0510 · 2026-07-06T08:12:16Z

Summary

This PR updates AionCore's aionrs provider resolution so custom providers use the selected model protocol instead of inferring protocol behavior from platform/
base URL.

Changes

Keep built-in platform defaults unchanged:
- anthropic -> Anthropic
- bedrock -> Bedrock
- gemini / openai -> OpenAI-compatible
- gemini-vertex-ai -> Vertex
For other providers, resolve protocol from provider.model_protocols[model_id].
Default missing or empty model_protocols to OpenAI-compatible.
Treat base_url as a user-provided endpoint prefix:
- OpenAI-compatible appends /chat/completions
- Anthropic-compatible appends /v1/messages
Stop stripping or special-casing /v1 suffixes from base_url.
Surface invalid model_protocols JSON as a bad request instead of silently falling back.
Apply the same provider/protocol resolution to provider health checks.

Why

The previous logic mixed protocol selection with platform/base URL assumptions. For custom providers such as DeepSeek, GLM, Poe, MiniMax, or InfiniAI, the
protocol should come from the user's selected model protocol. If no protocol is configured, the system should default to OpenAI-compatible behavior.

Validation

cargo test -p aionui-ai-agent resolve_custom_anthropic
cargo test -p aionui-ai-agent resolve_anthropic
cargo fmt --all -- --check
just push
- 6652 tests run: 6652 passed, 18 skipped

tcp404

crates/aionui-ai-agent/src/factory/aionrs.rs:L260: risk: OpenAI-compatible prefix handling looks right, but Anthropic/custom compatibility is incomplete. aionrs v0.1.38 AnthropicTransport still hardcodes format!("{base_url}/v1/messages") and does not use compat.api_path, so a stored base_url ending in /v1 regresses to /v1/v1/messages. Please convert these tests to table-driven cases and add a final URL semantics regression before merging.

Suggested table-driven cases:

Provider Mapping

platform	model_id	model_protocols	expected provider	expected result
`anthropic`	`m`	`None`	`anthropic`	ok
`bedrock`	`m`	`None`	`bedrock`	ok
`gemini`	`m`	`None`	`openai`	ok
`openai`	`m`	`None`	`openai`	ok
`gemini-vertex-ai`	`m`	`None`	`vertex`	ok
`custom`	`gpt-4o`	`None`	`openai`	ok
`new-api`	`m`	`None`	`openai`	ok
`unknown`	`m`	`None`	`openai`	ok
`new-api`	`claude`	`{"claude":"anthropic"}`	`anthropic`	ok
`new-api`	`gpt`	`{"gpt":"openai"}`	`openai`	ok
`new-api`	`gemini`	`{"gemini":"gemini"}`	`openai`	ok
`custom`	`claude`	`{"claude":"anthropic"}`	`anthropic`	ok
`custom`	`gpt`	`{"gpt":"openai"}`	`openai`	ok
`custom`	`unknown-model`	`{"claude":"anthropic"}`	`openai`	ok
`custom`	`m`	`{}`	`openai`	ok
`custom`	`m`	`{"m":""}`	`openai`	ok
`custom`	`m`	`{"m":123}`	`openai`	ok
`custom`	`m`	`not json`	-	error
`custom`	`m`	`{"m":"unsupported"}`	-	error
`openai`	`claude`	`{"claude":"anthropic"}`	`openai`	ok; known platform wins
`anthropic`	`gpt`	`{"gpt":"openai"}`	`anthropic`	ok; known platform wins

URL / Compat Resolution

platform	raw_base_url	mapped_provider	is_full_url	expected base_url	expected api_path	expected max_tokens_field
`custom`	`https://api.openai.com`	`openai`	false	same	`/chat/completions`	`max_completion_tokens`
`custom`	`https://api.openai.com/v1`	`openai`	false	`https://api.openai.com/v1`	`/chat/completions`	`max_completion_tokens`
`custom`	`https://api.openai.com/v1/`	`openai`	false	`https://api.openai.com/v1`	`/chat/completions`	`max_completion_tokens`
`custom`	`https://api.deepseek.com/v1`	`openai`	false	same	`/chat/completions`	none
`custom`	`https://open.bigmodel.cn/api/paas/v4`	`openai`	false	same	`/chat/completions`	none
`new-api`	`https://host/v1`	`openai`	false	same	`/chat/completions`	none
`custom`	`http://localhost:11434/v1`	`openai`	false	same	`/chat/completions`	none
`gemini`	`https://generativelanguage.googleapis.com`	`openai`	false	`https://generativelanguage.googleapis.com/v1beta/openai`	`/chat/completions`	none
`anthropic`	`https://api.anthropic.com`	`anthropic`	false	same	none	none
`anthropic`	`https://api.anthropic.com/`	`anthropic`	false	`https://api.anthropic.com`	none	none
`custom`	`https://proxy.example.com/anthropic`	`anthropic`	false	same	`/v1/messages` if final transport supports it	none
`custom`	`https://proxy.example.com/v1`	`anthropic`	false	`https://proxy.example.com` or final URL must not duplicate `/v1`	depends on chosen fix	none
`new-api`	`https://proxy.example.com/v1`	`anthropic`	false	`https://proxy.example.com` or final URL must not duplicate `/v1`	depends on chosen fix	none
`custom`	`https://proxy.example.com/v1/chat/completions`	`openai`	true	same without trailing slash	`""`	none
`custom`	`https://proxy.example.com/v1/messages`	`anthropic`	true	same without trailing slash	`""`	none
`custom`	``	`openai`	false	none	none	none

Final URL Semantics

These are the important regressions; testing only resolve_aionrs_url_and_compat() is not enough because Anthropic currently ignores compat.api_path.

provider	base_url passed to aionrs	compat.api_path	aionrs transport behavior	expected final URL
`openai`	`https://api.deepseek.com/v1`	`/chat/completions`	`base_url + api_path`	`https://api.deepseek.com/v1/chat/completions`
`openai`	`https://open.bigmodel.cn/api/paas/v4`	`/chat/completions`	`base_url + api_path`	`https://open.bigmodel.cn/api/paas/v4/chat/completions`
`openai`	`https://x/v1/chat/completions`	`""`	`base_url + ""`	`https://x/v1/chat/completions`
`anthropic`	`https://api.anthropic.com`	none	fixed `{base}/v1/messages`	`https://api.anthropic.com/v1/messages`
`anthropic`	`https://proxy.example.com`	maybe none	fixed `{base}/v1/messages`	`https://proxy.example.com/v1/messages`
`anthropic`	`https://proxy.example.com/v1`	currently `/v1/messages` in this PR	fixed `{base}/v1/messages`	must not be `https://proxy.example.com/v1/v1/messages`
`anthropic`	`https://x/v1/messages`	`""`	current aionrs still fixed-appends	verify/fix full-url mode for Anthropic too

jiahe0510 · 2026-07-06T12:41:25Z

crates/aionui-ai-agent/src/factory/aionrs.rs:L260: risk: OpenAI-compatible prefix handling looks right, but Anthropic/custom compatibility is incomplete. aionrs v0.1.38 AnthropicTransport still hardcodes format!("{base_url}/v1/messages") and does not use compat.api_path, so a stored base_url ending in /v1 regresses to /v1/v1/messages. Please convert these tests to table-driven cases and add a final URL semantics regression before merging.

Suggested table-driven cases:

Provider Mapping

platform model_id model_protocols expected provider expected result
anthropic m None anthropic ok
bedrock m None bedrock ok
gemini m None openai ok
openai m None openai ok
gemini-vertex-ai m None vertex ok
custom gpt-4o None openai ok
new-api m None openai ok
unknown m None openai ok
new-api claude {"claude":"anthropic"} anthropic ok
new-api gpt {"gpt":"openai"} openai ok
new-api gemini {"gemini":"gemini"} openai ok
custom claude {"claude":"anthropic"} anthropic ok
custom gpt {"gpt":"openai"} openai ok
custom unknown-model {"claude":"anthropic"} openai ok
custom m {} openai ok
custom m {"m":""} openai ok
custom m {"m":123} openai ok
custom m not json - error
custom m {"m":"unsupported"} - error
openai claude {"claude":"anthropic"} openai ok; known platform wins
anthropic gpt {"gpt":"openai"} anthropic ok; known platform wins

URL / Compat Resolution

platform raw_base_url mapped_provider is_full_url expected base_url expected api_path expected max_tokens_field
custom https://api.openai.com openai false same /chat/completions max_completion_tokens
custom https://api.openai.com/v1 openai false https://api.openai.com/v1 /chat/completions max_completion_tokens
custom https://api.openai.com/v1/ openai false https://api.openai.com/v1 /chat/completions max_completion_tokens
custom https://api.deepseek.com/v1 openai false same /chat/completions none
custom https://open.bigmodel.cn/api/paas/v4 openai false same /chat/completions none
new-api https://host/v1 openai false same /chat/completions none
custom http://localhost:11434/v1 openai false same /chat/completions none
gemini https://generativelanguage.googleapis.com openai false https://generativelanguage.googleapis.com/v1beta/openai /chat/completions none
anthropic https://api.anthropic.com anthropic false same none none
anthropic https://api.anthropic.com/ anthropic false https://api.anthropic.com none none
custom https://proxy.example.com/anthropic anthropic false same /v1/messages if final transport supports it none
custom https://proxy.example.com/v1 anthropic false https://proxy.example.com or final URL must not duplicate /v1 depends on chosen fix none
new-api https://proxy.example.com/v1 anthropic false https://proxy.example.com or final URL must not duplicate /v1 depends on chosen fix none
custom https://proxy.example.com/v1/chat/completions openai true same without trailing slash "" none
custom https://proxy.example.com/v1/messages anthropic true same without trailing slash "" none
custom `` openai false none none none

Final URL Semantics

These are the important regressions; testing only resolve_aionrs_url_and_compat() is not enough because Anthropic currently ignores compat.api_path.

provider base_url passed to aionrs compat.api_path aionrs transport behavior expected final URL
openai https://api.deepseek.com/v1 /chat/completions base_url + api_path https://api.deepseek.com/v1/chat/completions
openai https://open.bigmodel.cn/api/paas/v4 /chat/completions base_url + api_path https://open.bigmodel.cn/api/paas/v4/chat/completions
openai https://x/v1/chat/completions "" base_url + "" https://x/v1/chat/completions
anthropic https://api.anthropic.com none fixed {base}/v1/messages https://api.anthropic.com/v1/messages
anthropic https://proxy.example.com maybe none fixed {base}/v1/messages https://proxy.example.com/v1/messages
anthropic https://proxy.example.com/v1 currently /v1/messages in this PR fixed {base}/v1/messages must not be https://proxy.example.com/v1/v1/messages
anthropic https://x/v1/messages "" current aionrs still fixed-appends verify/fix full-url mode for Anthropic too

done

jiahe0510 added 5 commits July 6, 2026 14:06

fix(aionrs): default missing model protocol to openai

be519dc

fix(aionrs): respect model protocol over base url

10f3ce9

fix(aionrs): use versioned anthropic path for custom providers

e300395

fix(aionrs): stop inferring anthropic v1 suffix

ed8f26c

chore: restore cargo wrappers

3da228f

jiahe0510 requested review from kaizhou-lab, piorpua and tcp404 July 6, 2026 09:57

tcp404 requested changes Jul 6, 2026

View reviewed changes

jiahe0510 added 2 commits July 6, 2026 20:16

test(aionrs): cover model protocol url resolution

54ccf47

test(aionrs): cover preset provider urls

4b78b89

This comment was marked as duplicate.

Sign in to view

jiahe0510 closed this Jul 6, 2026

jiahe0510 reopened this Jul 6, 2026

jiahe0510 requested a review from tcp404 July 6, 2026 12:42

tcp404 approved these changes Jul 6, 2026

View reviewed changes

jiahe0510 removed request for kaizhou-lab and piorpua July 6, 2026 12:48

jiahe0510 merged commit 37f76a8 into main Jul 6, 2026
12 checks passed

jiahe0510 deleted the fix/model-protocol-url branch July 6, 2026 17:09

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(aionrs): default missing model protocol to openai#577

fix(aionrs): default missing model protocol to openai#577
jiahe0510 merged 7 commits into
mainfrom
fix/model-protocol-url

jiahe0510 commented Jul 6, 2026 •

edited

Loading

Uh oh!

tcp404 left a comment

Uh oh!

This comment was marked as duplicate.

Uh oh!

jiahe0510 commented Jul 6, 2026

Provider Mapping

URL / Compat Resolution

Final URL Semantics

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

jiahe0510 commented Jul 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Why

Validation

Uh oh!

tcp404 left a comment

Choose a reason for hiding this comment

Provider Mapping

URL / Compat Resolution

Final URL Semantics

Uh oh!

This comment was marked as duplicate.

Uh oh!

jiahe0510 commented Jul 6, 2026

Provider Mapping

URL / Compat Resolution

Final URL Semantics

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

jiahe0510 commented Jul 6, 2026 •

edited

Loading