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
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
|
import mime from "mime-types";
import {
Chat,
createPartFromBase64,
createPartFromUri,
createUserContent,
GoogleGenAI,
type Content,
type GeneratedImage,
type GeneratedVideo,
type Part,
} from "@google/genai";
import type { AIModelAPI, InputToken } from "./types";
import type { AsyncRes, Result } from "sortug";
export default class GeminiAPI implements AIModelAPI {
tokenizer: (text: string) => number;
maxTokens: number;
private model: string;
api: GoogleGenAI;
chats: Map<string, Chat> = new Map<string, Chat>();
constructor(
model?: string,
maxTokens = 200_000,
tokenizer: (text: string) => number = (text) => text.length / 3,
) {
this.maxTokens = maxTokens;
this.tokenizer = tokenizer;
const gem = new GoogleGenAI({ apiKey: Bun.env["GEMINI_API_KEY"]! });
this.api = gem;
this.model = model || "gemini-2.5-pro";
}
// input data in gemini gets pretty involved
//
// data
// Union type
// data can be only one of the following:
// text
// string
// Inline text.
// inlineData
// object (Blob)
// Inline media bytes.
// functionCall
// object (FunctionCall)
// A predicted FunctionCall returned from the model that contains a string representing the FunctionDeclaration.name with the arguments and their values.
// functionResponse
// object (FunctionResponse)
// The result output of a FunctionCall that contains a string representing the FunctionDeclaration.name and a structured JSON object containing any output from the function is used as context to the model.
// fileData
// object (FileData)
// URI based data.
// executableCode
// object (ExecutableCode)
// Code generated by the model that is meant to be executed.
// codeExecutionResult
// object (CodeExecutionResult)
// Result of executing the ExecutableCode.
// metadata
// Union type
public setModel(model: string) {
this.model = model;
}
private contentFromImage(imageString: string): Result<Part> {
// TODO
const mimeType = mime.lookup(imageString);
if (!mimeType) return { error: "no mimetype" };
const url = URL.parse(imageString);
if (url) {
const part = createPartFromUri(imageString, mimeType);
return { ok: part };
} else return { ok: createPartFromBase64(imageString, mimeType) };
}
public buildInput(tokens: InputToken[]): Result<Content> {
try {
const input = createUserContent(
tokens.map((t) => {
if ("text" in t) return t.text;
if ("img" in t) {
const imagePart = this.contentFromImage(t.img);
if ("error" in imagePart) throw new Error("image failed");
else return imagePart.ok;
}
return "oy vey";
}),
);
return { ok: input };
} catch (e) {
return { error: `${e}` };
}
}
async send(input: string | Content, systemPrompt?: string): AsyncRes<string> {
try {
const opts = {
model: this.model,
contents: input,
};
const fopts = systemPrompt
? { ...opts, config: { systemInstruction: systemPrompt } }
: opts;
const response = await this.api.models.generateContent(fopts);
if (!response.text) return { error: "no text in response" };
return { ok: response.text };
} catch (e) {
return { error: `${e}` };
}
}
async stream(
input: string | Content,
handler: (s: string) => void,
systemPrompt?: string,
) {
const opts = {
model: this.model,
contents: input,
};
const fopts = systemPrompt
? { ...opts, config: { systemInstruction: systemPrompt } }
: opts;
const response = await this.api.models.generateContentStream(fopts);
for await (const chunk of response) {
handler(chunk.text || "");
}
}
async makeImage(prompt: string): AsyncRes<GeneratedImage[]> {
try {
const response = await this.api.models.generateImages({
model: this.model,
prompt,
});
// TODO if empty or undefined return error
return { ok: response.generatedImages || [] };
} catch (e) {
return { error: `${e}` };
}
}
async makeVideo({
prompt,
image,
}: {
prompt?: string;
image?: string;
}): AsyncRes<GeneratedVideo[]> {
try {
const response = await this.api.models.generateVideos({
model: this.model,
prompt,
});
// TODO if empty or undefined return error
return { ok: response.response?.generatedVideos || [] };
} catch (e) {
return { error: `${e}` };
}
}
}
// TODO how to use caches
// https://ai.google.dev/api/caching
|
