Expand description
OpenAI-style tool-call emulation.
Venice E2EE responses do not expose native function calls, so this module
parses tool calls client-side from decrypted assistant text using a vendored
subset of vLLM’s Rust tool parser, validates the calls against the request’s OpenAI
tools, and builds prompt text for the encrypted controller/correction
requests.
Venice E2EE cannot render tools into the server-side chat template (tools arrive encrypted), so the proxy prompt must instruct a model-visible output shape. Use model-family-specific shapes where the bundled vLLM Rust parser has a matching format, and fall back to Hermes JSON otherwise.
Structs§
- Tool
Call Validation Error - Validation error for parsed assistant tool-call output.
- Tool
Emulation Context - Per-request tool-emulation state derived from config, tools, and tool choice.
- Validated
Tool Call - OpenAI-compatible tool call validated against the request’s available tools.
Enums§
- Tool
Output Classification - Classification of a decrypted assistant response under tool emulation.
Functions§
- generate_
tool_ call_ id - Generates an OpenAI-style tool-call ID.