[jiahuei]

Summary

• Added Molto.embeddings() with a typed OpenAI request and parsed SDK result, native OpenAI-compatible transport for OpenAI/Infinity/vLLM, and Cohere v2/VoyageAI text embedding translation. Endpoint, provider, and destination remain independent; self-hosted routes require an explicit URL.
• Preserved single/batched text and supported token inputs, vector indices, dimensions, SDK encoding defaults/overrides, native task/truncation options, model identity, and provider metadata. Unknown usage remains None. Unsupported multimodal/contextual, quantized, sparse, and multiple-vector semantics fail explicitly. Shared HTTP lifecycle, credentials, header filtering, and error conventions are reused without new runtime dependencies, streaming, or automatic batching.
• Added independent hosted recordings and clearly labelled synthetic schema fixtures with public-coordinator replay coverage. Existing conversational cassettes and all original embedding recordings remain unchanged. A newly added synthetic Infinity error enum was corrected to the pinned schema's Empty value.
• Published the five-provider support matrix, examples, architecture, and evidence limits. This PR is independent of feat: Add provider clients and multi-turn replay coverage #96.

Draft; related to #93. OpenAI (text-embedding-3-small) and Cohere (embed-v4.0) have successful live captures. VoyageAI's available credentials returned 401; the configured vLLM server timed out; Infinity was unconfigured. VoyageAI success coverage is synthetic. Infinity 3.0.0 and vLLM 0.33.0 are pinned source/schema references, not live-tested server/model claims. The issue remains open until successful evidence establishes all five integrations.

Verification

• uv run --extra test pytest -q --tb=short: 1,625 passed; replay only, including existing conversational regressions.
• uv run --extra test pytest -q --tb=short tests/test_embedding_translation.py tests/core/test_embeddings.py: 108 passed.
• uv run --extra test pytest -q --tb=short tests/core/test_embedding_engines.py: 56 passed; corrected synthetic error also passed focused replay.
• uv run --extra lint ruff check --output-format github --config pyproject.toml .: passed.
• uv run --extra lint ruff format --diff --config pyproject.toml .: passed, 100 files.
• npx prettier@3.8 --check .: passed; final matrix wording also passed a focused check.
• git diff --check and recorded-fixture integrity checks: passed.
• gh pr checks 97 --repo EmbeddedLLM/molto --watch --interval 10: all checks passed, including Python 3.12, 3.13, and 3.14, Ruff, and Prettier.
• Live verification remains incomplete for VoyageAI, Infinity, and vLLM for the access/deployment reasons above.