feat: synchronize fork with upstream v0.6.0-alpha.2 by AlexMikhalev · Pull Request #4 · terraphim/rust-genai

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev · 2026-02-07T11:30:27Z

Summary

Merge 124 upstream commits from jeremychone/rust-genai v0.6.0-alpha.2
Preserve all fork-specific adapters (Bedrock, Cerebras, OpenRouter, Zhipu)
Integrate new upstream features (ModelSpec, MIMO, BigModel, enhanced ZAI)
Fix API changes and resolve 12 merge conflicts

Changes

Upstream Features Integrated

ModelSpec API: Enhanced model specification system
MIMO Adapter: New provider support
BigModel Adapter: bigmodel.cn provider support
Enhanced ZAI: Dual endpoint support (zai::, zai-coding::), renamed to "Zai"
Cache Control TTL: Anthropic cache with 5m/1h TTL options

Fork Features Preserved

AWS Bedrock: Converse API adapter with Bearer token auth
Cerebras: Fast inference provider
OpenRouter: Multi-provider routing
Zhipu: Legacy compatibility

Adapters (19 total)

OpenAI, OpenAIResp, Gemini, OpenRouter, Anthropic, Fireworks, Together, Groq, Mimo, Nebius, Xai, DeepSeek, Zai, BigModel, Cohere, Ollama, Cerebras, Bedrock, Zhipu

Test plan

cargo fmt --check - passes
cargo clippy - 6 warnings (all in upstream code)
cargo build - compiles successfully
cargo test --lib - 14/14 tests pass
cargo test --test mock_tests - 18/18 tests pass
cargo test --test test_adapter_consistency - verifies adapter code matches tests
cargo test --test test_verify_model_lists - verifies model resolution
cargo test --test test_zai_adapter - verifies ZAI adapter

Defects Resolved

Issue	Resolution
Method `as_model_name_and_namespace` renamed	Updated to `namespace_and_name()`
Duplicate `size()` methods	Removed duplicates
ZAI adapter renamed "ZAi" -> "Zai"	Updated all tests
ZAI model list changed	Updated test expectations

Generated by Terraphim AI

…one#94)

…ave the fn_arguments parsed

…eremychone#95) - Rename zhipu adapter to zai adapter - Implement namespace-based routing for regular vs coding endpoints - Regular models: "glm-4.6" → credit-based API - Coding models: "zai::glm-4.6" → subscription API - Add comprehensive example c07-zai.rs - Update adapter kind detection to handle "zai" namespace - Remove dual-endpoint complexity from user-facing API Co-authored-by: Rui Andrada <rui.andrada@imaginedone.com.br>

(cherry picked from commit 448e6ec)

…file in git)

… adapter)

* feat: parts thought signature * feat: gemini adapter thought signature support - Add ContentPart::ThoughtSignature with accessors and helpers. - Introduce InterStreamEvent::ThoughtSignatureChunk and ChatStreamEvent::ThoughtSignatureChunk. - Capture thought signatures into StreamEnd.captured_content when enabled. - Update printer to display ThoughtSignatureChunk events (non-captured). - Gemini adapter: parse thoughtSignature (fallback: thought) and include in outgoing messages. - Gemini streamer: surface thought as ThoughtSignatureChunk and record in captured_thought_signatures. - Preserve thoughtSignature in assistant history before tool calls when looping tools. - Example c10-tooluse-streaming.rs: switch to gemini-3-pro-preview, log thoughtSignature chunks, and prepend captured thoughts to assistant message. * chore: impl thought signature * chore: no API change

…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

… payload)

…hone#137)

…s not given (jeremychone#137)

…ng API calls (jeremychone#126)

- fix - anthropic - streaming adapter now captures cache tokens ^ cache - anthropic - Adds them to prompt_tokens (matching into_usage behavior) ^ cache - anthropic - add TTL support for prompt caching - fix - anthropic - apply per-part cache_control on system messages and add cache TTL example

Merged 124 upstream commits from jeremychone/rust-genai while preserving fork-specific adapters: - AWS Bedrock (Converse API with Bearer token auth) - Cerebras (OpenAI-compatible) - OpenRouter (provider-pattern routing) - Zhipu (legacy, backwards compatibility) New features from upstream: - ModelSpec API for flexible model specification - MIMO and BigModel adapters - Enhanced ZAI with dual-endpoint support (zai:: and zai-coding::) - ThoughtSignature and ReasoningEffort improvements - Custom EventSourceStream replacing reqwest-eventsource - Enhanced error handling with request/response capture Breaking changes: - ZAi adapter renamed to Zai (upstream naming) - ZAI model list updated to upstream v0.6.0-alpha.2 Conflict resolution: - adapter_kind.rs: Merged all 19 adapter variants - dispatcher.rs: Added fork adapters to upstream dispatch - adapters/mod.rs: Export all adapter modules - Accepted upstream for streaming and chat improvements Tests updated for new adapter naming and model lists. Co-Authored-By: Terraphim AI <noreply@terraphim.io>

Apply cargo fmt to fix minor formatting inconsistencies in: - src/chat/tool/tool_response.rs (extra blank line) - tests/test_verify_model_lists.rs (println formatting) - tests/test_zai_adapter.rs (import order) Co-Authored-By: Terraphim AI <noreply@terraphim.io>

…pproach

Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

pdylanrossand others added 30 commits October 19, 2025 18:23

added ModelName partial eq implementations for string types (jeremych…
6b3d3e0
…one#94)

Fixed streaming tool calls for openai models (jeremychone#91)
d3fcf0b

- stream tool - openai - fix issue that captured_tool_calls ado not h…
4123ee2
…ave the fn_arguments parsed

. anthropic - update model name for haiku 4.5
4aef5ca

. anthropic - update claude-haiku(4.5) max tokens to 64k as per spec
d234513

. first pass at adding the module spec
db19d5d

. cargo fmt
28011fd

. update to v0.4.3
1181667

. v0.4.4-WIP
7b30e42

. gitignore - add audio
545a8c0

. gitignore - add ignore doc types
95783b4

Add support for OpenAIservice_tierparameter (jeremychone#98)
cb24f4f

Add support for audio_type content part for voice agent support.
039678f
(cherry picked from commit 448e6ec)

. openai audio - vgami commit about adding test (jc edit: minus .wav …
66ff7fc
…file in git)

. minor typo & clippy
c17b9ea

+ openai - adding support for gpt-5-pro (must be mapped to OpenaiResp…
9022f1a
… adapter)

. test - openai - added gpt-5.1
caa8395

. openai - update model names
0dc9e28

. readme and changelog
41d172a

. update to v0.4.4
176a05a

. update to v0.4.5-WIP
b0bf4da

Merge branch 'pr-ii64-gemini-thought-signature'
5f9bdc3

. update 0.5.0-WIP
2fd9df9

. clippy clean
5ff9a40

. anthropic - update with opus-4-5 (and set correct max_tokens)
c654781

. gemini model names
abee836

. test - fix wikipedia image issue with image url (use aipack image url)
7652866

^ gemini - for gemini-3, convert ReasoningEffort Low/High to the appr…
01783e3
…opriate gemini thinkingLevel LOW/HIGH, fall back on budget if not gemini 3 or other effort

jeremychoneand others added 26 commits January 27, 2026 18:55

. version 0.5.2
f19d08c

. 0.5.3-WIP
bda3edb

. ChatRole - Add PartialEq / Eq (jeremychone#131)
8e28018

Fix incorrect empty output from MessageContent::joined_texts for ≥ 2 …
9682697
…text parts (fixesjeremychone#135) (jeremychone#136) Co-authored-by: Ross MacLeod <rmm+github@z.odi.ac>

. tracing - add traced to web-client for ai response (jeremychone#132)
b403a55

. gemini test - add test_tool_google_web_search_ok
0b2366e

. test - use flash 3 for test_tool_google_web_search_ok
fa72baa

. minor refact - webc post new take & of payload (did not need to own…
6bba9f0
… payload)

> refactor captured_raw_body into client .exec_chat (prep forjeremyc…
b68f0de
…hone#137)

^ error - add request payload / response body when to chat response fail
295d8cd

. error - add comment
ce57146

^ error - ChatResponseGeneration - add 'doc text' when raw response i…
1dd352a
…s not given (jeremychone#137)

. v0.5.3
9a66c28

. 0.5.4-WIP
c9720c3

. cargo fmt
b134187

. 0.6.0-alpha.1-WIP
21b634c

feat: Add ModelSpec to allow for additional model details when maki…
c3d4b97
…ng API calls (jeremychone#126)

^ ModelName - add Static/Shared inner
a177724

. ModelSpec - update ModelSpec from apis
bdec7a5

+ example - c06-model-spec
de31e44

^ ModelSpec - implement missing From ModelName
2491f61

. v0.6.0-alpha.1
4b83c8e

. 0.6.0-alpha.2-WIP
b3fe812

AlexMikhalev merged commit b6ba1b3 into mainFeb 7, 2026
1 check failed

AlexMikhalev added a commit that referenced this pull request Mar 16, 2026
symphony:#4- Reconcile OpenRouter adapter with upstream namespace a…
3877796
…pproach

AlexMikhalev added a commit that referenced this pull request Mar 16, 2026
fix: resolve merge conflict in CLAUDE.md for issue#4
3cc3e38
Refs #4 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

AlexMikhalev added a commit that referenced this pull request Mar 16, 2026
Merge#4: Reconcile OpenRouter adapter with upstream namespace approach
ac249a6