vtcode-llm 0.177.0

LLM provider abstraction, client implementations, and streaming for VT Code
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75

//! Chat Completions and Responses request builders for OpenAI-compatible APIs.
//!
//! Model policy and Responses history shaping live in private child modules.

use vtcode_config::core::OpenAIHostedShellConfig;
use vtcode_config::types::ReasoningEffortLevel;

mod chat;
mod history;
mod model_policy;
mod responses;

pub(crate) use chat::build_chat_request;
pub(crate) use responses::build_responses_request;

use model_policy::{is_gpt6_model, is_gpt56_model};

pub(super) fn supports_prompt_cache_options(model: &str) -> bool {
    is_gpt56_model(model) || is_gpt6_model(model)
}

pub(crate) struct ChatRequestContext<'a> {
    pub model: &'a str,
    pub is_native_openai: bool,
    pub supports_tools: bool,
    pub supports_parallel_tool_config: bool,
    pub supports_temperature: bool,
    pub prompt_cache_key: Option<&'a str>,
    pub default_service_tier: Option<&'a str>,
}

pub(crate) struct ResponsesRequestContext<'a> {
    pub supports_tools: bool,
    pub supports_allowed_tools: bool,
    pub supports_parallel_tool_config: bool,
    pub supports_temperature: bool,
    pub supports_reasoning_effort: bool,
    pub supported_reasoning_efforts: &'a [&'a str],
    pub supports_reasoning: bool,
    pub is_responses_api_model: bool,
    pub include_max_output_tokens: bool,
    pub include_output_types: bool,
    pub include_sampling_parameters: bool,
    pub force_response_store_false: bool,
    pub include_assistant_phase: bool,
    pub prompt_cache_key: Option<&'a str>,
    pub include_prompt_cache_retention: bool,
    pub prompt_cache_retention: Option<&'a str>,
    /// Emit explicit `prompt_cache_breakpoint` markers on the stable prefix.
    /// Only native api.openai.com honors them: third-party
    /// Responses-compatible endpoints (e.g. Merge Gateway via a base-URL
    /// override) reject the field with 400 `invalid_parameter`, and the
    /// ChatGPT subscription backend rejects it with 400
    /// `prompt_cache_breakpoint is not supported on this model` (observed
    /// live on `gpt-6.1-sol`). Implicit breakpoints still cover the latest
    /// message on those backends.
    pub include_explicit_cache_breakpoints: bool,
    pub default_service_tier: Option<&'a str>,
    pub default_response_store: Option<bool>,
    pub default_responses_include: Option<&'a [String]>,
    pub include_encrypted_reasoning: bool,
    pub hosted_shell: Option<&'a OpenAIHostedShellConfig>,
    pub include_structured_history_in_input: bool,
    pub preserve_structured_history_on_replay: bool,
    pub preserve_assistant_phase_on_replay: bool,
    pub reasoning_context: Option<&'a str>,
    pub safety_identifier: Option<&'a str>,
}

pub(super) fn trimmed_non_empty(value: Option<&str>) -> Option<&str> {
    value.map(str::trim).filter(|value| !value.is_empty())
}

#[cfg(test)]
mod tests;