fix(anthropic): complete third-party Anthropic-compatible provider support by teknium1 · Pull Request #12846 · NousResearch/hermes-agent

teknium1 · 2026-04-20T05:28:09Z

Summary

Third-party Anthropic-compatible gateways (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com: auto-detected transport, no OAuth leak, prompt caching on, auxiliary tasks routed correctly, truncation continuation supported, and stale config cleared on provider switch.

Synthesizes eight stale community PRs into one consolidated change. Tests are passing across tests/run_agent, tests/agent, tests/hermes_cli (4554 passed, 0 regressions).

What this fixes, at a glance

Broken on main	Now
`base_url` ending in `/anthropic` sometimes falls through to `chat_completions`	One helper `_detect_api_mode_for_url` covers all three runtime-resolution paths
`_is_anthropic_oauth` can be `True` on MiniMax/GLM — injects Claude-Code identity headers → 401/403	All 5 assignment sites gate on `provider == "anthropic"`
Prompt caching only on native Anthropic + OpenRouter	Also on third-party Anthropic gateways (`api_mode=anthropic_messages` + Claude) with correct layout selection
`native_anthropic=True` sent to proxies that need envelope layout	New `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)`
Auxiliary tasks (compression, web_extract) on a custom Anthropic endpoint silently downgrade to OpenAI-wire client	`_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient`
Switching from custom MiniMax → OpenRouter leaves stale `api_mode`/`api_key` in config.yaml	`_update_config_for_provider` clears them
Anthropic `stop_reason: max_tokens` returns half-finished response, no retry	Continuation loop (text) + tool-call retry both cover `anthropic_messages`

Changes

hermes_cli/runtime_provider.py — _detect_api_mode_for_url now also maps */anthropic → anthropic_messages. The three inline endswith("/anthropic") duplicates in _resolve_runtime_from_pool_entry / explicit-provider path / api-key-provider path all route through it.

run_agent.py —

New method _anthropic_prompt_cache_policy(provider, base_url, api_mode, model) -> (bool, bool): single source of truth for "should we cache?" + "native or envelope layout?". Callers: __init__, switch_model, _try_activate_fallback.
New attribute _use_native_cache_layout persists layout choice. Threaded through _primary_runtime save/restore (with a backward-compat default for snapshots that predate the attribute).
All 5 _is_oauth_token() assignments guarded with if <native-provider-check> else False.
Truncation continuation: the gate if self.api_mode in ("chat_completions", "bedrock_converse"): now includes "anthropic_messages". The Anthropic response is normalized once via normalize_anthropic_response() at the top of the truncation block, so text-continuation and tool-call-retry branches share one code path for all three api_modes.

agent/auxiliary_client.py — _try_custom_endpoint gains an anthropic_messages branch that builds AnthropicAuxiliaryClient (is_oauth=False). Falls back to OpenAI-wire + logged warning when the anthropic SDK is unavailable.

hermes_cli/auth.py — _update_config_for_provider clears stale api_key/api_mode when switching to a built-in provider.

scripts/release.py — AUTHOR_MAP entries for 7 new contributors.

Validation

	Before	After
`/anthropic` detection locations	3 inline duplicates + helper	1 helper, 3 callers
OAuth guard coverage	2/5 sites	5/5 sites
Caching on third-party Anthropic gateway	Off	On (native layout)
`anthropic_messages` truncation	Drops response, no retry	Retries up to 3× like other api_modes
Aux client on custom anthropic endpoint	Wrong wire protocol	Native Anthropic wrapper
Targeted test run	—	42 new tests pass
Full run (`run_agent`/`agent`/`hermes_cli`)	—	4554 passed, 0 regressions

Synthesized from

Good parts of each PR were cherry-picked into the final design. Contributor authorship preserved via Co-authored-by: trailers.

fix: auto-detect anthropic_messages api_mode for third-party /anthropic endpoints #7410 @nocoo — URL-detection helper (kept + extended to consolidate 3 inline duplicates)
fix(agent): prevent identity injection for third-party Anthropic providers #7393 @keyuyuan — OAuth 5-site guard pattern (kept)
fix(agent): prevent false OAuth detection for third-party Anthropic providers #7367 @n-WN — OAuth guard explanatory comment (kept; mechanism subsumed by fix(agent): prevent identity injection for third-party Anthropic providers #7393)
fix(agent): enable prompt caching on Anthropic-compatible endpoints #8636 @sgaofen — caching helper factoring + native-vs-proxy layout split (kept; simplified from 2 helpers returning bools to 1 helper returning a tuple)
fix: enable prompt caching for Anthropic proxy endpoints #10954 @Only-Code-A — caching on anthropic_messages + Claude model names (kept)
fix(aux): honor api_mode for custom auxiliary endpoints #7648 @zhongyueming1121 — aux client anthropic_messages branch (kept; passes is_oauth=False for third-party safety)
fix(cli): persist api_mode when switching to named custom provider #6096 @hansnow — _update_config_for_provider stale-field clear (kept the auth.py hunk; _model_flow_named_custom hunk was already on main, skipped)
fix: add truncation continuation support for anthropic_messages API mode #9691 @TroyMitchell911 — truncation continuation (kept; simplified to one normalization call at the top of the truncation block so both retry paths share it)

Explicitly rejected

fix: enable prompt caching for Claude on any OpenAI-compatible endpoint #9621 (caching on OpenAI-wire via blocklist) — incomplete blocklist (missing Azure, Mistral, Fireworks) would cause 400s on strict-schema providers that reject unknown cache_control keys. Not worth the risk.
fix(agent): add length-continuation support for anthropic_messages mode #7242 — superseded by fix: add truncation continuation support for anthropic_messages API mode #9691 (duplicate handler, more invasive, references variables incorrectly against current main layout).
fix(routing): pass api_mode from cheap_model config in smart routing #8321 — targets agent/smart_model_routing.py which was removed in refactor: remove smart_model_routing feature #12732; the feature no longer exists.

PRs to close after this merges

Close with credit:

fix: auto-detect anthropic_messages api_mode for third-party /anthropic endpoints #7410, fix(agent): prevent false OAuth detection for third-party Anthropic providers #7367, fix(agent): prevent identity injection for third-party Anthropic providers #7393, fix(agent): enable prompt caching on Anthropic-compatible endpoints #8636, fix: enable prompt caching for Anthropic proxy endpoints #10954, fix(aux): honor api_mode for custom auxiliary endpoints #7648, fix(cli): persist api_mode when switching to named custom provider #6096, fix: add truncation continuation support for anthropic_messages API mode #9691

Close as rejected (with reasoning):

fix: enable prompt caching for Claude on any OpenAI-compatible endpoint #9621, fix(agent): add length-continuation support for anthropic_messages mode #7242, fix(routing): pass api_mode from cheap_model config in smart routing #8321

Plus two linked issues to close:

[Bug]: Third-party Anthropic-compatible providers get Claude Code identity injected #7366 (third-party Anthropic identity-injection bug report)
[Bug]: Prompt caching is not enabled for custom Anthropic-compatible providers using api_mode: anthropic_messages #8294 (prompt caching not enabled for custom Anthropic-compatible providers)

@nocoo

…pport Third-party gateways that speak the native Anthropic protocol (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com callers. Synthesizes eight stale community PRs into one consolidated change. Five fixes: - URL detection: consolidate three inline `endswith("/anthropic")` checks in runtime_provider.py into the shared _detect_api_mode_for_url helper. Third-party /anthropic endpoints now auto-resolve to api_mode=anthropic_messages via one code path instead of three. - OAuth leak-guard: all five sites that assign `_is_anthropic_oauth` (__init__, switch_model, _try_refresh_anthropic_client_credentials, _swap_credential, _try_activate_fallback) now gate on `provider == "anthropic"` so a stale ANTHROPIC_TOKEN never trips Claude-Code identity injection on third-party endpoints. Previously only 2 of 5 sites were guarded. - Prompt caching: new method `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)` per endpoint. Replaces three inline conditions and the `native_anthropic=(api_mode=='anthropic_messages')` call-site flag. Native Anthropic and third-party Anthropic gateways both get the native cache_control layout; OpenRouter gets envelope layout. Layout is persisted in `_primary_runtime` so fallback restoration preserves the per-endpoint choice. - Auxiliary client: `_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient` instead of silently downgrading to an OpenAI-wire client. Degrades gracefully to OpenAI-wire when the anthropic SDK isn't installed. - Config hygiene: `_update_config_for_provider` (hermes_cli/auth.py) clears stale `api_key`/`api_mode` when switching to a built-in provider, so a previous MiniMax custom endpoint's credentials can't leak into a later OpenRouter session. - Truncation continuation: length-continuation and tool-call-truncation retry now cover `anthropic_messages` in addition to `chat_completions` and `bedrock_converse`. Reuses the existing `_build_assistant_message` path via `normalize_anthropic_response()` so the interim message shape is byte-identical to the non-truncated path. Tests: 6 new files, 42 test cases. Targeted run + tests/run_agent, tests/agent, tests/hermes_cli all pass (4554 passed). Synthesized from (credits preserved via Co-authored-by trailers): #7410 @nocoo — URL detection helper #7393 @keyuyuan — OAuth 5-site guard #7367 @n-WN — OAuth guard (narrower cousin, kept comment) #8636 @sgaofen — caching helper + native-vs-proxy layout split #10954 @Only-Code-A — caching on anthropic_messages+Claude #7648 @zhongyueming1121 — aux client anthropic_messages branch #6096 @hansnow — /model switch clears stale api_mode #9691 @TroyMitchell911 — anthropic_messages truncation continuation Closes: #7366, #8294 (third-party Anthropic identity + caching). Supersedes: #7410, #7367, #7393, #8636, #10954, #7648, #6096, #9691. Rejects: #9621 (OpenAI-wire caching with incomplete blocklist — risky), #7242 (superseded by #9691, stale branch), #8321 (targets smart_model_routing which was removed in #12732). Co-authored-by: nocoo <nocoo@users.noreply.github.com> Co-authored-by: Keyu Yuan <leoyuan0099@gmail.com> Co-authored-by: Zoee <30841158+n-WN@users.noreply.github.com> Co-authored-by: sgaofen <135070653+sgaofen@users.noreply.github.com> Co-authored-by: Only-Code-A <bxzt2006@163.com> Co-authored-by: zhongyueming <mygamez@163.com> Co-authored-by: Xiaohan Li <hansnow@users.noreply.github.com> Co-authored-by: Troy Mitchell <i@troy-y.org>

…_messages transport PR NousResearch#12846 enabled Anthropic prompt caching for third-party gateways, but gated it on is_claude, which excluded providers like MiniMax that serve their own model families (MiniMax-M2.7, etc.) through the native Anthropic protocol. MiniMax documents full cache_control support on its /anthropic endpoints (global and China). This patch adds MiniMax detection to _anthropic_prompt_cache_policy() using: - Built-in provider id (minimax, minimax-cn), or - Known Anthropic-compatible hostname (api.minimax.io, api.minimaxi.com) Both paths receive the native cache_control layout. Refs: NousResearch#8294 (related, but only covered Claude-named models on third-party gateways). Closes NousResearch#17332

#17425) MiniMax's /anthropic endpoint documents cache_control support (0.1x read pricing, 5-min TTL) for MiniMax-M2.7, M2.5, M2.1, M2. PR #12846 gated third-party Anthropic-wire caching on 'claude' in model name, which left MiniMax's own model family re-paying full input tokens every turn. Opt in explicitly via provider id (minimax / minimax-cn) or host match (api.minimax.io / api.minimaxi.com). Narrow allowlist mirroring the existing Qwen/Alibaba branch below; leaves room for a capability-based surface (ProviderConfig.supports_anthropic_cache) if a third provider needs it. Closes #17332

@nocoo

…pport (NousResearch#12846) Third-party gateways that speak the native Anthropic protocol (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com callers. Synthesizes eight stale community PRs into one consolidated change. Five fixes: - URL detection: consolidate three inline `endswith("/anthropic")` checks in runtime_provider.py into the shared _detect_api_mode_for_url helper. Third-party /anthropic endpoints now auto-resolve to api_mode=anthropic_messages via one code path instead of three. - OAuth leak-guard: all five sites that assign `_is_anthropic_oauth` (__init__, switch_model, _try_refresh_anthropic_client_credentials, _swap_credential, _try_activate_fallback) now gate on `provider == "anthropic"` so a stale ANTHROPIC_TOKEN never trips Claude-Code identity injection on third-party endpoints. Previously only 2 of 5 sites were guarded. - Prompt caching: new method `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)` per endpoint. Replaces three inline conditions and the `native_anthropic=(api_mode=='anthropic_messages')` call-site flag. Native Anthropic and third-party Anthropic gateways both get the native cache_control layout; OpenRouter gets envelope layout. Layout is persisted in `_primary_runtime` so fallback restoration preserves the per-endpoint choice. - Auxiliary client: `_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient` instead of silently downgrading to an OpenAI-wire client. Degrades gracefully to OpenAI-wire when the anthropic SDK isn't installed. - Config hygiene: `_update_config_for_provider` (hermes_cli/auth.py) clears stale `api_key`/`api_mode` when switching to a built-in provider, so a previous MiniMax custom endpoint's credentials can't leak into a later OpenRouter session. - Truncation continuation: length-continuation and tool-call-truncation retry now cover `anthropic_messages` in addition to `chat_completions` and `bedrock_converse`. Reuses the existing `_build_assistant_message` path via `normalize_anthropic_response()` so the interim message shape is byte-identical to the non-truncated path. Tests: 6 new files, 42 test cases. Targeted run + tests/run_agent, tests/agent, tests/hermes_cli all pass (4554 passed). Synthesized from (credits preserved via Co-authored-by trailers): NousResearch#7410 @nocoo — URL detection helper NousResearch#7393 @keyuyuan — OAuth 5-site guard NousResearch#7367 @n-WN — OAuth guard (narrower cousin, kept comment) NousResearch#8636 @sgaofen — caching helper + native-vs-proxy layout split NousResearch#10954 @Only-Code-A — caching on anthropic_messages+Claude NousResearch#7648 @zhongyueming1121 — aux client anthropic_messages branch NousResearch#6096 @hansnow — /model switch clears stale api_mode NousResearch#9691 @TroyMitchell911 — anthropic_messages truncation continuation Closes: NousResearch#7366, NousResearch#8294 (third-party Anthropic identity + caching). Supersedes: NousResearch#7410, NousResearch#7367, NousResearch#7393, NousResearch#8636, NousResearch#10954, NousResearch#7648, NousResearch#6096, NousResearch#9691. Rejects: NousResearch#9621 (OpenAI-wire caching with incomplete blocklist — risky), NousResearch#7242 (superseded by NousResearch#9691, stale branch), NousResearch#8321 (targets smart_model_routing which was removed in NousResearch#12732). Co-authored-by: nocoo <nocoo@users.noreply.github.com> Co-authored-by: Keyu Yuan <leoyuan0099@gmail.com> Co-authored-by: Zoee <30841158+n-WN@users.noreply.github.com> Co-authored-by: sgaofen <135070653+sgaofen@users.noreply.github.com> Co-authored-by: Only-Code-A <bxzt2006@163.com> Co-authored-by: zhongyueming <mygamez@163.com> Co-authored-by: Xiaohan Li <hansnow@users.noreply.github.com> Co-authored-by: Troy Mitchell <i@troy-y.org>

@nocoo

…pport (NousResearch#12846) Third-party gateways that speak the native Anthropic protocol (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com callers. Synthesizes eight stale community PRs into one consolidated change. Five fixes: - URL detection: consolidate three inline `endswith("/anthropic")` checks in runtime_provider.py into the shared _detect_api_mode_for_url helper. Third-party /anthropic endpoints now auto-resolve to api_mode=anthropic_messages via one code path instead of three. - OAuth leak-guard: all five sites that assign `_is_anthropic_oauth` (__init__, switch_model, _try_refresh_anthropic_client_credentials, _swap_credential, _try_activate_fallback) now gate on `provider == "anthropic"` so a stale ANTHROPIC_TOKEN never trips Claude-Code identity injection on third-party endpoints. Previously only 2 of 5 sites were guarded. - Prompt caching: new method `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)` per endpoint. Replaces three inline conditions and the `native_anthropic=(api_mode=='anthropic_messages')` call-site flag. Native Anthropic and third-party Anthropic gateways both get the native cache_control layout; OpenRouter gets envelope layout. Layout is persisted in `_primary_runtime` so fallback restoration preserves the per-endpoint choice. - Auxiliary client: `_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient` instead of silently downgrading to an OpenAI-wire client. Degrades gracefully to OpenAI-wire when the anthropic SDK isn't installed. - Config hygiene: `_update_config_for_provider` (hermes_cli/auth.py) clears stale `api_key`/`api_mode` when switching to a built-in provider, so a previous MiniMax custom endpoint's credentials can't leak into a later OpenRouter session. - Truncation continuation: length-continuation and tool-call-truncation retry now cover `anthropic_messages` in addition to `chat_completions` and `bedrock_converse`. Reuses the existing `_build_assistant_message` path via `normalize_anthropic_response()` so the interim message shape is byte-identical to the non-truncated path. Tests: 6 new files, 42 test cases. Targeted run + tests/run_agent, tests/agent, tests/hermes_cli all pass (4554 passed). Synthesized from (credits preserved via Co-authored-by trailers): NousResearch#7410 @nocoo — URL detection helper NousResearch#7393 @keyuyuan — OAuth 5-site guard NousResearch#7367 @n-WN — OAuth guard (narrower cousin, kept comment) NousResearch#8636 @sgaofen — caching helper + native-vs-proxy layout split NousResearch#10954 @Only-Code-A — caching on anthropic_messages+Claude NousResearch#7648 @zhongyueming1121 — aux client anthropic_messages branch NousResearch#6096 @hansnow — /model switch clears stale api_mode NousResearch#9691 @TroyMitchell911 — anthropic_messages truncation continuation Closes: NousResearch#7366, NousResearch#8294 (third-party Anthropic identity + caching). Supersedes: NousResearch#7410, NousResearch#7367, NousResearch#7393, NousResearch#8636, NousResearch#10954, NousResearch#7648, NousResearch#6096, NousResearch#9691. Rejects: NousResearch#9621 (OpenAI-wire caching with incomplete blocklist — risky), NousResearch#7242 (superseded by NousResearch#9691, stale branch), NousResearch#8321 (targets smart_model_routing which was removed in NousResearch#12732). Co-authored-by: nocoo <nocoo@users.noreply.github.com> Co-authored-by: Keyu Yuan <leoyuan0099@gmail.com> Co-authored-by: Zoee <30841158+n-WN@users.noreply.github.com> Co-authored-by: sgaofen <135070653+sgaofen@users.noreply.github.com> Co-authored-by: Only-Code-A <bxzt2006@163.com> Co-authored-by: zhongyueming <mygamez@163.com> Co-authored-by: Xiaohan Li <hansnow@users.noreply.github.com> Co-authored-by: Troy Mitchell <i@troy-y.org>

@nocoo

…pport (NousResearch#12846) Third-party gateways that speak the native Anthropic protocol (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com callers. Synthesizes eight stale community PRs into one consolidated change. Five fixes: - URL detection: consolidate three inline `endswith("/anthropic")` checks in runtime_provider.py into the shared _detect_api_mode_for_url helper. Third-party /anthropic endpoints now auto-resolve to api_mode=anthropic_messages via one code path instead of three. - OAuth leak-guard: all five sites that assign `_is_anthropic_oauth` (__init__, switch_model, _try_refresh_anthropic_client_credentials, _swap_credential, _try_activate_fallback) now gate on `provider == "anthropic"` so a stale ANTHROPIC_TOKEN never trips Claude-Code identity injection on third-party endpoints. Previously only 2 of 5 sites were guarded. - Prompt caching: new method `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)` per endpoint. Replaces three inline conditions and the `native_anthropic=(api_mode=='anthropic_messages')` call-site flag. Native Anthropic and third-party Anthropic gateways both get the native cache_control layout; OpenRouter gets envelope layout. Layout is persisted in `_primary_runtime` so fallback restoration preserves the per-endpoint choice. - Auxiliary client: `_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient` instead of silently downgrading to an OpenAI-wire client. Degrades gracefully to OpenAI-wire when the anthropic SDK isn't installed. - Config hygiene: `_update_config_for_provider` (hermes_cli/auth.py) clears stale `api_key`/`api_mode` when switching to a built-in provider, so a previous MiniMax custom endpoint's credentials can't leak into a later OpenRouter session. - Truncation continuation: length-continuation and tool-call-truncation retry now cover `anthropic_messages` in addition to `chat_completions` and `bedrock_converse`. Reuses the existing `_build_assistant_message` path via `normalize_anthropic_response()` so the interim message shape is byte-identical to the non-truncated path. Tests: 6 new files, 42 test cases. Targeted run + tests/run_agent, tests/agent, tests/hermes_cli all pass (4554 passed). Synthesized from (credits preserved via Co-authored-by trailers): NousResearch#7410 @nocoo — URL detection helper NousResearch#7393 @keyuyuan — OAuth 5-site guard NousResearch#7367 @n-WN — OAuth guard (narrower cousin, kept comment) NousResearch#8636 @sgaofen — caching helper + native-vs-proxy layout split NousResearch#10954 @Only-Code-A — caching on anthropic_messages+Claude NousResearch#7648 @zhongyueming1121 — aux client anthropic_messages branch NousResearch#6096 @hansnow — /model switch clears stale api_mode NousResearch#9691 @TroyMitchell911 — anthropic_messages truncation continuation Closes: NousResearch#7366, NousResearch#8294 (third-party Anthropic identity + caching). Supersedes: NousResearch#7410, NousResearch#7367, NousResearch#7393, NousResearch#8636, NousResearch#10954, NousResearch#7648, NousResearch#6096, NousResearch#9691. Rejects: NousResearch#9621 (OpenAI-wire caching with incomplete blocklist — risky), NousResearch#7242 (superseded by NousResearch#9691, stale branch), NousResearch#8321 (targets smart_model_routing which was removed in NousResearch#12732). Co-authored-by: nocoo <nocoo@users.noreply.github.com> Co-authored-by: Keyu Yuan <leoyuan0099@gmail.com> Co-authored-by: Zoee <30841158+n-WN@users.noreply.github.com> Co-authored-by: sgaofen <135070653+sgaofen@users.noreply.github.com> Co-authored-by: Only-Code-A <bxzt2006@163.com> Co-authored-by: zhongyueming <mygamez@163.com> Co-authored-by: Xiaohan Li <hansnow@users.noreply.github.com> Co-authored-by: Troy Mitchell <i@troy-y.org>

NousResearch#17425) MiniMax's /anthropic endpoint documents cache_control support (0.1x read pricing, 5-min TTL) for MiniMax-M2.7, M2.5, M2.1, M2. PR NousResearch#12846 gated third-party Anthropic-wire caching on 'claude' in model name, which left MiniMax's own model family re-paying full input tokens every turn. Opt in explicitly via provider id (minimax / minimax-cn) or host match (api.minimax.io / api.minimaxi.com). Narrow allowlist mirroring the existing Qwen/Alibaba branch below; leaves room for a capability-based surface (ProviderConfig.supports_anthropic_cache) if a third provider needs it. Closes NousResearch#17332

ydawei · 2026-05-04T00:56:11Z

Observing this for the Gemini native provider. Is this fix needed for the provider too?

@nocoo

…pport (NousResearch#12846) Third-party gateways that speak the native Anthropic protocol (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com callers. Synthesizes eight stale community PRs into one consolidated change. Five fixes: - URL detection: consolidate three inline `endswith("/anthropic")` checks in runtime_provider.py into the shared _detect_api_mode_for_url helper. Third-party /anthropic endpoints now auto-resolve to api_mode=anthropic_messages via one code path instead of three. - OAuth leak-guard: all five sites that assign `_is_anthropic_oauth` (__init__, switch_model, _try_refresh_anthropic_client_credentials, _swap_credential, _try_activate_fallback) now gate on `provider == "anthropic"` so a stale ANTHROPIC_TOKEN never trips Claude-Code identity injection on third-party endpoints. Previously only 2 of 5 sites were guarded. - Prompt caching: new method `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)` per endpoint. Replaces three inline conditions and the `native_anthropic=(api_mode=='anthropic_messages')` call-site flag. Native Anthropic and third-party Anthropic gateways both get the native cache_control layout; OpenRouter gets envelope layout. Layout is persisted in `_primary_runtime` so fallback restoration preserves the per-endpoint choice. - Auxiliary client: `_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient` instead of silently downgrading to an OpenAI-wire client. Degrades gracefully to OpenAI-wire when the anthropic SDK isn't installed. - Config hygiene: `_update_config_for_provider` (hermes_cli/auth.py) clears stale `api_key`/`api_mode` when switching to a built-in provider, so a previous MiniMax custom endpoint's credentials can't leak into a later OpenRouter session. - Truncation continuation: length-continuation and tool-call-truncation retry now cover `anthropic_messages` in addition to `chat_completions` and `bedrock_converse`. Reuses the existing `_build_assistant_message` path via `normalize_anthropic_response()` so the interim message shape is byte-identical to the non-truncated path. Tests: 6 new files, 42 test cases. Targeted run + tests/run_agent, tests/agent, tests/hermes_cli all pass (4554 passed). Synthesized from (credits preserved via Co-authored-by trailers): NousResearch#7410 @nocoo — URL detection helper NousResearch#7393 @keyuyuan — OAuth 5-site guard NousResearch#7367 @n-WN — OAuth guard (narrower cousin, kept comment) NousResearch#8636 @sgaofen — caching helper + native-vs-proxy layout split NousResearch#10954 @Only-Code-A — caching on anthropic_messages+Claude NousResearch#7648 @zhongyueming1121 — aux client anthropic_messages branch NousResearch#6096 @hansnow — /model switch clears stale api_mode NousResearch#9691 @TroyMitchell911 — anthropic_messages truncation continuation Closes: NousResearch#7366, NousResearch#8294 (third-party Anthropic identity + caching). Supersedes: NousResearch#7410, NousResearch#7367, NousResearch#7393, NousResearch#8636, NousResearch#10954, NousResearch#7648, NousResearch#6096, NousResearch#9691. Rejects: NousResearch#9621 (OpenAI-wire caching with incomplete blocklist — risky), NousResearch#7242 (superseded by NousResearch#9691, stale branch), NousResearch#8321 (targets smart_model_routing which was removed in NousResearch#12732). Co-authored-by: nocoo <nocoo@users.noreply.github.com> Co-authored-by: Keyu Yuan <leoyuan0099@gmail.com> Co-authored-by: Zoee <30841158+n-WN@users.noreply.github.com> Co-authored-by: sgaofen <135070653+sgaofen@users.noreply.github.com> Co-authored-by: Only-Code-A <bxzt2006@163.com> Co-authored-by: zhongyueming <mygamez@163.com> Co-authored-by: Xiaohan Li <hansnow@users.noreply.github.com> Co-authored-by: Troy Mitchell <i@troy-y.org>

NousResearch#17425) MiniMax's /anthropic endpoint documents cache_control support (0.1x read pricing, 5-min TTL) for MiniMax-M2.7, M2.5, M2.1, M2. PR NousResearch#12846 gated third-party Anthropic-wire caching on 'claude' in model name, which left MiniMax's own model family re-paying full input tokens every turn. Opt in explicitly via provider id (minimax / minimax-cn) or host match (api.minimax.io / api.minimaxi.com). Narrow allowlist mirroring the existing Qwen/Alibaba branch below; leaves room for a capability-based surface (ProviderConfig.supports_anthropic_cache) if a third provider needs it. Closes NousResearch#17332

@nocoo

…pport (NousResearch#12846) Third-party gateways that speak the native Anthropic protocol (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com callers. Synthesizes eight stale community PRs into one consolidated change. Five fixes: - URL detection: consolidate three inline `endswith("/anthropic")` checks in runtime_provider.py into the shared _detect_api_mode_for_url helper. Third-party /anthropic endpoints now auto-resolve to api_mode=anthropic_messages via one code path instead of three. - OAuth leak-guard: all five sites that assign `_is_anthropic_oauth` (__init__, switch_model, _try_refresh_anthropic_client_credentials, _swap_credential, _try_activate_fallback) now gate on `provider == "anthropic"` so a stale ANTHROPIC_TOKEN never trips Claude-Code identity injection on third-party endpoints. Previously only 2 of 5 sites were guarded. - Prompt caching: new method `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)` per endpoint. Replaces three inline conditions and the `native_anthropic=(api_mode=='anthropic_messages')` call-site flag. Native Anthropic and third-party Anthropic gateways both get the native cache_control layout; OpenRouter gets envelope layout. Layout is persisted in `_primary_runtime` so fallback restoration preserves the per-endpoint choice. - Auxiliary client: `_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient` instead of silently downgrading to an OpenAI-wire client. Degrades gracefully to OpenAI-wire when the anthropic SDK isn't installed. - Config hygiene: `_update_config_for_provider` (hermes_cli/auth.py) clears stale `api_key`/`api_mode` when switching to a built-in provider, so a previous MiniMax custom endpoint's credentials can't leak into a later OpenRouter session. - Truncation continuation: length-continuation and tool-call-truncation retry now cover `anthropic_messages` in addition to `chat_completions` and `bedrock_converse`. Reuses the existing `_build_assistant_message` path via `normalize_anthropic_response()` so the interim message shape is byte-identical to the non-truncated path. Tests: 6 new files, 42 test cases. Targeted run + tests/run_agent, tests/agent, tests/hermes_cli all pass (4554 passed). Synthesized from (credits preserved via Co-authored-by trailers): NousResearch#7410 @nocoo — URL detection helper NousResearch#7393 @keyuyuan — OAuth 5-site guard NousResearch#7367 @n-WN — OAuth guard (narrower cousin, kept comment) NousResearch#8636 @sgaofen — caching helper + native-vs-proxy layout split NousResearch#10954 @Only-Code-A — caching on anthropic_messages+Claude NousResearch#7648 @zhongyueming1121 — aux client anthropic_messages branch NousResearch#6096 @hansnow — /model switch clears stale api_mode NousResearch#9691 @TroyMitchell911 — anthropic_messages truncation continuation Closes: NousResearch#7366, NousResearch#8294 (third-party Anthropic identity + caching). Supersedes: NousResearch#7410, NousResearch#7367, NousResearch#7393, NousResearch#8636, NousResearch#10954, NousResearch#7648, NousResearch#6096, NousResearch#9691. Rejects: NousResearch#9621 (OpenAI-wire caching with incomplete blocklist — risky), NousResearch#7242 (superseded by NousResearch#9691, stale branch), NousResearch#8321 (targets smart_model_routing which was removed in NousResearch#12732). Co-authored-by: nocoo <nocoo@users.noreply.github.com> Co-authored-by: Keyu Yuan <leoyuan0099@gmail.com> Co-authored-by: Zoee <30841158+n-WN@users.noreply.github.com> Co-authored-by: sgaofen <135070653+sgaofen@users.noreply.github.com> Co-authored-by: Only-Code-A <bxzt2006@163.com> Co-authored-by: zhongyueming <mygamez@163.com> Co-authored-by: Xiaohan Li <hansnow@users.noreply.github.com> Co-authored-by: Troy Mitchell <i@troy-y.org>

NousResearch#17425) MiniMax's /anthropic endpoint documents cache_control support (0.1x read pricing, 5-min TTL) for MiniMax-M2.7, M2.5, M2.1, M2. PR NousResearch#12846 gated third-party Anthropic-wire caching on 'claude' in model name, which left MiniMax's own model family re-paying full input tokens every turn. Opt in explicitly via provider id (minimax / minimax-cn) or host match (api.minimax.io / api.minimaxi.com). Narrow allowlist mirroring the existing Qwen/Alibaba branch below; leaves room for a capability-based surface (ProviderConfig.supports_anthropic_cache) if a third provider needs it. Closes NousResearch#17332

@nocoo

…pport (NousResearch#12846) Third-party gateways that speak the native Anthropic protocol (MiniMax, Zhipu GLM, Alibaba DashScope, Kimi, LiteLLM proxies) now work end-to-end with the same feature set as direct api.anthropic.com callers. Synthesizes eight stale community PRs into one consolidated change. Five fixes: - URL detection: consolidate three inline `endswith("/anthropic")` checks in runtime_provider.py into the shared _detect_api_mode_for_url helper. Third-party /anthropic endpoints now auto-resolve to api_mode=anthropic_messages via one code path instead of three. - OAuth leak-guard: all five sites that assign `_is_anthropic_oauth` (__init__, switch_model, _try_refresh_anthropic_client_credentials, _swap_credential, _try_activate_fallback) now gate on `provider == "anthropic"` so a stale ANTHROPIC_TOKEN never trips Claude-Code identity injection on third-party endpoints. Previously only 2 of 5 sites were guarded. - Prompt caching: new method `_anthropic_prompt_cache_policy()` returns `(should_cache, use_native_layout)` per endpoint. Replaces three inline conditions and the `native_anthropic=(api_mode=='anthropic_messages')` call-site flag. Native Anthropic and third-party Anthropic gateways both get the native cache_control layout; OpenRouter gets envelope layout. Layout is persisted in `_primary_runtime` so fallback restoration preserves the per-endpoint choice. - Auxiliary client: `_try_custom_endpoint` honors `api_mode=anthropic_messages` and builds `AnthropicAuxiliaryClient` instead of silently downgrading to an OpenAI-wire client. Degrades gracefully to OpenAI-wire when the anthropic SDK isn't installed. - Config hygiene: `_update_config_for_provider` (hermes_cli/auth.py) clears stale `api_key`/`api_mode` when switching to a built-in provider, so a previous MiniMax custom endpoint's credentials can't leak into a later OpenRouter session. - Truncation continuation: length-continuation and tool-call-truncation retry now cover `anthropic_messages` in addition to `chat_completions` and `bedrock_converse`. Reuses the existing `_build_assistant_message` path via `normalize_anthropic_response()` so the interim message shape is byte-identical to the non-truncated path. Tests: 6 new files, 42 test cases. Targeted run + tests/run_agent, tests/agent, tests/hermes_cli all pass (4554 passed). Synthesized from (credits preserved via Co-authored-by trailers): NousResearch#7410 @nocoo — URL detection helper NousResearch#7393 @keyuyuan — OAuth 5-site guard NousResearch#7367 @n-WN — OAuth guard (narrower cousin, kept comment) NousResearch#8636 @sgaofen — caching helper + native-vs-proxy layout split NousResearch#10954 @Only-Code-A — caching on anthropic_messages+Claude NousResearch#7648 @zhongyueming1121 — aux client anthropic_messages branch NousResearch#6096 @hansnow — /model switch clears stale api_mode NousResearch#9691 @TroyMitchell911 — anthropic_messages truncation continuation Closes: NousResearch#7366, NousResearch#8294 (third-party Anthropic identity + caching). Supersedes: NousResearch#7410, NousResearch#7367, NousResearch#7393, NousResearch#8636, NousResearch#10954, NousResearch#7648, NousResearch#6096, NousResearch#9691. Rejects: NousResearch#9621 (OpenAI-wire caching with incomplete blocklist — risky), NousResearch#7242 (superseded by NousResearch#9691, stale branch), NousResearch#8321 (targets smart_model_routing which was removed in NousResearch#12732). Co-authored-by: nocoo <nocoo@users.noreply.github.com> Co-authored-by: Keyu Yuan <leoyuan0099@gmail.com> Co-authored-by: Zoee <30841158+n-WN@users.noreply.github.com> Co-authored-by: sgaofen <135070653+sgaofen@users.noreply.github.com> Co-authored-by: Only-Code-A <bxzt2006@163.com> Co-authored-by: zhongyueming <mygamez@163.com> Co-authored-by: Xiaohan Li <hansnow@users.noreply.github.com> Co-authored-by: Troy Mitchell <i@troy-y.org>

NousResearch#17425) MiniMax's /anthropic endpoint documents cache_control support (0.1x read pricing, 5-min TTL) for MiniMax-M2.7, M2.5, M2.1, M2. PR NousResearch#12846 gated third-party Anthropic-wire caching on 'claude' in model name, which left MiniMax's own model family re-paying full input tokens every turn. Opt in explicitly via provider id (minimax / minimax-cn) or host match (api.minimax.io / api.minimaxi.com). Narrow allowlist mirroring the existing Qwen/Alibaba branch below; leaves room for a capability-based surface (ProviderConfig.supports_anthropic_cache) if a third provider needs it. Closes NousResearch#17332

teknium1 merged commit 65a31ee into main Apr 20, 2026
4 of 6 checks passed

teknium1 deleted the fix/third-party-anthropic-messages-unified branch April 20, 2026 05:43

eggressive mentioned this pull request Apr 21, 2026

[Feature]: Enable Anthropic prompt caching for Claude models on Nous Portal #13409

Open

1 task

EurFelux mentioned this pull request Apr 26, 2026

[Bug]: 'custom = OpenAI' hardcoded assumption: explicit_base_url path ignores api_mode (sibling of #7661/#15033, plus key_env never parsed in aux config) #16254

Closed

nicoechaniz mentioned this pull request Apr 29, 2026

[Bug]: Prompt caching silently disabled for MiniMax's own models (MiniMax-M2.7 etc.) on anthropic_messages transport #17332

Closed

nicoechaniz mentioned this pull request Apr 29, 2026

fix(agent): enable prompt caching for MiniMax own models on anthropic_messages transport #17333

Closed

teknium1 mentioned this pull request Apr 29, 2026

fix(minimax): enable Anthropic prompt caching for MiniMax's own models #17425

Merged

rodboev mentioned this pull request Jun 2, 2026

[Bug]: Smart routing drops api_mode from cheap_model config #8515

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(anthropic): complete third-party Anthropic-compatible provider support#12846

fix(anthropic): complete third-party Anthropic-compatible provider support#12846
teknium1 merged 1 commit into
mainfrom
fix/third-party-anthropic-messages-unified

teknium1 commented Apr 20, 2026

Uh oh!

Uh oh!

ydawei commented May 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

teknium1 commented Apr 20, 2026

Summary

What this fixes, at a glance

Changes

Validation

Synthesized from

Explicitly rejected

PRs to close after this merges

Uh oh!

Uh oh!

ydawei commented May 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants