Skip to main content

Phase 14. LLM bindings

FieldValue
MEPMEP-45 §Phases · Phase 14
StatusNOT STARTED
Started
Landed
Tracking issue
Tracking PR

Gate

LLM fixture suite (~20 cases: generate, embed, chat against OpenAI/Anthropic/Google/llama.cpp) compiles + runs byte-equal vs vm3 in replay mode (recorded cassettes); live-mode runs available behind a flag.

Goal-alignment audit

To be written before sub-phase 14.0 starts. LLM surface is the AI-augmented workflow Mochi positions itself for; without it, the language is missing a marketed feature in native builds. Aligns.

Sub-phases

#ScopeStatusCommitPR
14.0Provider abstraction in transpiler3/c/runtime/src/llm/: trait-like vtable per providerNOT STARTED
14.1OpenAI provider (libcurl + yyjson)NOT STARTED
14.2Anthropic providerNOT STARTED
14.3Google providerNOT STARTED
14.4llama.cpp local provider (linked only with --with-llama)NOT STARTED
14.5Replay-mode cassettes under tests/transpiler3/c/cassettes/llm/; cassette layer intercepts libcurlNOT STARTED

Decisions made

Fill in along the way.

Deferred work

Provider-specific tool-use / function-calling integration tests: a follow-up phase if upstream stabilises.

Closeout notes

Fill in after gate green.