LOCAL / CLI SYNTHESIS — no Anthropic HTTP batch API.
Synthesis backends (v1.4.0+)
The planned Anthropic HTTP API mode (batch + prompt-cache backend) and the agent-delegate pending-prompt flow were not shipped as production backends. v1.4.0 supports:
| Backend | Config | Docs |
|---|---|---|
dummy |
default | offline stubs for tests |
ollama |
synthesis.backend: ollama |
Tutorial 08 |
claude |
synthesis.backend: claude |
Agent / Claude CLI |
Cost estimates still use the rate card in model_pricing.csv /
llmwiki synth --estimate. Prompt-caching helpers in
llmwiki/cache.py remain for estimate math; there is no
sync --batch or .llmwiki-batch-state.json path.
Setup (Claude CLI)
{
"synthesis": {
"backend": "claude",
"claude_model": "sonnet"
}
}
llmwiki synth --check
llmwiki synth --estimate
llmwiki synth