unigateway 1.3.0

Lightweight, local-first LLM gateway for developers. A stable, single-binary unified entry point for all your AI tools and models.
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
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152

use std::future::Future;
use std::sync::Arc;

use axum::response::Response;
use tracing::info;
use unigateway_host::{
    flow::{missing_upstream_api_key_response, resolve_core_only_host_flow},
    host::{HostEnvProvider, HostPoolSource},
};
use unigateway_protocol::RuntimeHttpResponse;

use crate::types::GatewayRequestState;

use super::request_flow::PreparedGatewayRequest;
use super::response_flow::respond_prepared_host_result;

type BeforeExecuteHook = for<'a> fn(&GatewayRequestState, &PreparedGatewayRequest<'a>);

#[derive(Clone, Copy)]
pub(super) struct HostExecutionSpec {
    endpoint: &'static str,
    env_provider: HostEnvProvider,
    unavailable_message: &'static str,
    before_execute: Option<BeforeExecuteHook>,
}

pub(super) fn openai_chat_spec() -> HostExecutionSpec {
    HostExecutionSpec {
        endpoint: "/v1/chat/completions",
        env_provider: HostEnvProvider::OpenAi,
        unavailable_message: "no provider pool available for chat",
        before_execute: None,
    }
}

pub(super) fn openai_responses_spec() -> HostExecutionSpec {
    HostExecutionSpec {
        endpoint: "/v1/responses",
        env_provider: HostEnvProvider::OpenAi,
        unavailable_message: "no openai-compatible provider available for responses",
        before_execute: None,
    }
}

pub(super) fn anthropic_chat_spec() -> HostExecutionSpec {
    HostExecutionSpec {
        endpoint: "/v1/messages",
        env_provider: HostEnvProvider::Anthropic,
        unavailable_message: "no provider pool available for messages",
        before_execute: Some(log_anthropic_request_execution),
    }
}

pub(super) fn openai_embeddings_spec() -> HostExecutionSpec {
    HostExecutionSpec {
        endpoint: "/v1/embeddings",
        env_provider: HostEnvProvider::OpenAi,
        unavailable_message: "no openai-compatible provider available for embeddings",
        before_execute: None,
    }
}

pub(super) async fn execute_prepared_host_request<'prepared, Request, Execute, CoreFuture>(
    state: &Arc<GatewayRequestState>,
    prepared: &'prepared PreparedGatewayRequest<'_>,
    request: Request,
    spec: HostExecutionSpec,
    execute: Execute,
) -> Response
where
    Execute: FnOnce(
        &'prepared PreparedGatewayRequest<'_>,
        HostPoolSource<'prepared>,
        Request,
    ) -> CoreFuture,
    CoreFuture: Future<Output = anyhow::Result<Option<RuntimeHttpResponse>>>,
{
    if let Some(before_execute) = spec.before_execute {
        before_execute(state.as_ref(), prepared);
    }

    let source = match host_pool_source(state, prepared, spec.env_provider) {
        Ok(source) => source,
        Err(HostPoolSourceError::MissingUpstreamApiKey) => {
            return respond_prepared_host_result(
                state,
                prepared,
                spec.endpoint,
                Err(missing_upstream_api_key_response()),
            )
            .await;
        }
    };

    let result =
        resolve_core_only_host_flow(execute(prepared, source, request), spec.unavailable_message)
            .await;
    respond_prepared_host_result(state, prepared, spec.endpoint, result).await
}

fn host_pool_source<'prepared>(
    state: &GatewayRequestState,
    prepared: &'prepared PreparedGatewayRequest<'_>,
    env_provider: HostEnvProvider,
) -> Result<HostPoolSource<'prepared>, HostPoolSourceError> {
    if let Some(auth) = prepared.auth.as_ref() {
        return Ok(HostPoolSource::Service(&auth.key.service_id));
    }

    let api_key_override = (!prepared.token.is_empty()).then_some(prepared.token.as_str());
    if api_key_override.is_none() && env_api_key(state, env_provider).is_empty() {
        return Err(HostPoolSourceError::MissingUpstreamApiKey);
    }

    Ok(HostPoolSource::Env {
        provider: env_provider,
        api_key_override,
    })
}

fn env_api_key(state: &GatewayRequestState, provider: HostEnvProvider) -> &str {
    state.provider_api_key(provider)
}

fn log_anthropic_request_execution(
    state: &GatewayRequestState,
    prepared: &PreparedGatewayRequest<'_>,
) {
    let endpoint = "/v1/messages";
    let env_api_key = env_api_key(state, HostEnvProvider::Anthropic);

    info!(
        endpoint,
        gateway_key_matched = prepared.auth.is_some(),
        token_present = !prepared.token.is_empty(),
        "anthropic request authentication result"
    );

    if prepared.auth.is_none() {
        info!(
            endpoint,
            token_present = !prepared.token.is_empty(),
            env_key_present = !env_api_key.is_empty(),
            using_env_fallback = prepared.token.is_empty() && !env_api_key.is_empty(),
            "anthropic request falling back to env upstream key"
        );
    }
}

enum HostPoolSourceError {
    MissingUpstreamApiKey,
}