21789375408

Committed 08 Feb 2026 12:27AM UTC coverage: 54.544% (-9.7%) from 64.266%

Build # 21789375408

Build Type

push

github

Committed by

web-flow

Commit Message

feat: add Antigravity and Zai provider (#24)

## Summary
This PR applies formatting improvements across the codebase and documentation, while adding support for the Antigravity provider with Gemini 3 Flash and Claude Sonnet 4.5 models.

## Key Changes

### Formatting & Code Style
- Fixed inconsistent spacing in JSON/YAML files (`.devcontainer/devcontainer.json`, `.github/workflows/`)
- Reformatted long lines in documentation files to improve readability (`.github/CHANGE_PROVIDER.md`, `.github/TOOL_USAGE.md`, `README.md`)
- Removed trailing whitespace and fixed line endings across multiple files
- Fixed trailing comma in `.vscode/launch.json` configuration

### Documentation Updates
- Updated `README.md` with comprehensive provider and model availability table
- Added detailed Antigravity provider configuration guide with OAuth2 setup instructions
- Improved formatting of environment variable documentation
- Updated default model reference from Phind 70B to Pollinations
- Enhanced `AGENTS.md` with better section spacing and organization

### New Feature: Antigravity Provider
- Added Antigravity provider support in `main.ts` with command handlers for:
  - `antigravity:` / `antigemini:` - Gemini 3 Flash
  - `anticlaude:` - Claude Sonnet 4.5
- Integrated with existing `TelegramService.callAntigravityModel` infrastructure
- Supports tool usage for both models

## Implementation Details
- Antigravity uses OAuth2 refresh token authentication with automatic access token renewal
- Project ID is auto-discovered via API if not explicitly configured
- Access tokens are valid for ~1 hour and refreshed automatically
- Configuration requires `ANTIGRAVITY_REFRESH_TOKEN` environment variable
- Optional `ANTIGRAVITY_PROJECT_ID` for explicit project specification

https://claude.ai/code/session_01XHVz46ba9iwhgeBAxtuMps

<!-- This is an auto-generated comment: release notes by coderabbit.ai -->
## Summary by CodeRabbit

* **New Features**
  * Suporte ao provedor Antigrav... (continued)

Run Details

185 of 374 branches covered (49.47%)

Branch coverage included in aggregate %.

477 of 1576 new or added lines in 17 files covered. (30.27%)

4 existing lines in 2 files now uncovered.

2864 of 5216 relevant lines covered (54.91%)

17.34 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

58.08

/src/prototype/ContextExtensionPrototype.ts

import { Context } from 'grammy';
import { Action } from 'grammy-auto-chat-action-types';
import { Audio, Message, ParseMode, PhotoSize, Voice } from 'grammy-types';
import { transcribeAudio } from '@/service/TelegramService.ts';
import { toTelegramMarkdown } from '@/util/MarkdownUtils.ts';

const MARKDOWN_ERROR_MESSAGE = 'Error on markdown parse_mode, message:';

declare module 'grammy' {
        interface Context {
                replyWithQuote(
                        output: string,
                        config?: { parse_mode: ParseMode },
                ): Promise<Message.TextMessage>;

                replyWithVisionNotSupportedByModel(): Promise<Message.TextMessage>;

                startTypingIndicator(): number;

                replyInChunks(output: string): void;

                streamReply(
                        reader: ReadableStreamDefaultReader<Uint8Array>,
                        onComplete: (completedAnswer: string) => Promise<void>,
                        responseMap?: (responseBody: string) => string,
                        lastResult?: string,
                ): Promise<void>;

                extractContextKeys(): Promise<{
                        userId: number;
                        userKey: string;
                        contextMessage?: string;
                        audio?: Voice | Audio;
                        photos?: PhotoSize[];
                        caption?: string;
                        quote?: string;
                }>;

                chatAction: Action | undefined;
        }
}

/**
 * Reply to a message with quoting the original message
 */
Context.prototype.replyWithQuote = function (
        this: Context,
        output: string,
        config?: { parse_mode: ParseMode },
) {
        return this.reply(output, {
                reply_to_message_id: this.message?.message_id,
                ...config,
        });
};

/**
 * Reply that the model doesn't support vision capabilities
 */
Context.prototype.replyWithVisionNotSupportedByModel = function (
        this: Context,
) {
        return this.replyWithQuote('esse modelo não suporta leitura de foto');
};

/**
 * Start typing indicator that persists by re-sending every 4 seconds
 */
Context.prototype.startTypingIndicator = function (this: Context): number {
        this.chatAction = 'typing';
        return setInterval(() => {
                this.chatAction = 'typing';
        }, 4000);
};

/**
 * Split a large response into multiple message chunks
 */
Context.prototype.replyInChunks = function (
        this: Context,
        output: string,
): void {
        if (output.length > 4096) {
                const outputChunks = output.match(/[\s\S]{1,4093}/g)!;

                outputChunks.forEach((chunk, index) => {
                        const isLastChunk = index === outputChunks.length - 1;
                        const chunkOutput = `${chunk}${isLastChunk ? '' : '...'}`;
                        const sanitizedOutput = toTelegramMarkdown(chunkOutput);

                        this.replyWithQuote(sanitizedOutput, { parse_mode: 'Markdown' })
                                .catch(() => {
                                        console.warn(MARKDOWN_ERROR_MESSAGE, chunkOutput);
                                        this.replyWithQuote(chunkOutput);
                                });
                });
                return;
        }

        const sanitizedOutput = toTelegramMarkdown(output);
        this.replyWithQuote(sanitizedOutput, { parse_mode: 'Markdown' })
                .catch(() => {
                        console.warn(MARKDOWN_ERROR_MESSAGE, output);
                        this.replyWithQuote(output);
                });
};

/**
 * Stream a response to the user with periodic updates
 */
Context.prototype.streamReply = async function (
        this: Context,
        reader: ReadableStreamDefaultReader<Uint8Array>,
        onComplete: (completedAnswer: string) => Promise<void>,
        responseMap?: (responseBody: string) => string,
        lastResult?: string,
): Promise<void> {
        const { message_id } = await this.replyWithQuote('processando...');
        let result = lastResult || '';
        let lastUpdate = Date.now();
        let lastSentMessage = '';

        while (true) {
                const { done, value } = await reader.read();
                if (done) break;

                const chunk = decodeStreamResponseText(value, responseMap);
                result += chunk;

                if (result.length > 4093) {
                        result = result.removeThinkingChatCompletion()
                                .convertBlackBoxWebSearchSourcesToMarkdown();

                        if (result.length > 4093) {
                                const remainingChunk = result.substring(4093) + chunk;
                                result = result.substring(0, 4093);

                                const updateResult = await editMessageWithCompletionEvery3Seconds(
                                        this,
                                        message_id,
                                        result,
                                        lastUpdate,
                                        lastSentMessage,
                                        true,
                                );
                                lastUpdate = updateResult.timestamp;
                                lastSentMessage = updateResult.lastMessage;
                                onComplete(result);
                                return this.streamReply(
                                        reader,
                                        onComplete,
                                        responseMap,
                                        remainingChunk,
                                );
                        }
                }

                const updateResult = await editMessageWithCompletionEvery3Seconds(
                        this,
                        message_id,
                        result,
                        lastUpdate,
                        lastSentMessage,
                );
                lastUpdate = updateResult.timestamp;
                lastSentMessage = updateResult.lastMessage;
        }

        let sanitizedResult = result.removeThinkingChatCompletion()
                .convertBlackBoxWebSearchSourcesToMarkdown();

        if (sanitizedResult.length > 4093) {
                const remainingChunk = sanitizedResult.substring(4093);
                sanitizedResult = sanitizedResult.substring(0, 4093) + '...';
                this.replyInChunks(remainingChunk);
        }

        this.api.editMessageText(this.chat!.id, message_id, toTelegramMarkdown(sanitizedResult), {
                parse_mode: 'Markdown',
        })
                .catch(() => {
                        console.warn(MARKDOWN_ERROR_MESSAGE, sanitizedResult);
                        this.api.editMessageText(this.chat!.id, message_id, sanitizedResult);
                });

        onComplete(result);
};

/**
 * Extract common context keys from the message
 */
Context.prototype.extractContextKeys = async function (this: Context) {
        const userId = this.from?.id!;
        const userKey = `user:${userId}`;
        const audio = this.message?.voice || this.message?.audio;
        const contextMessage = await getTextMessage(userId, userKey, this, audio);
        const photos = this.message?.photo;
        const caption = this.message?.caption;
        const quote = this.message?.reply_to_message?.text;

        return { userId, userKey, contextMessage, audio, photos, caption, quote };
};

/**
 * Helper function to get text from a message, transcribing audio if needed
 */
function getTextMessage(
        userId: number,
        userKey: string,
        ctx: Context,
        audio?: Voice,
): Promise<string | undefined> {
        if (audio) {
                return transcribeAudio(userId, userKey, ctx, audio);
        }
        return Promise.resolve(ctx.message?.text);
}

/**
 * Decode text from the response stream
 */
function decodeStreamResponseText(
        responseMessage: Uint8Array,
        responseMap?: (responseBody: string) => string,
): string {
        const decoder = new TextDecoder();
        const decodedText = decoder.decode(responseMessage);
        return responseMap ? responseMap(decodedText) : decodedText;
}

/**
 * Edit a message with updated content, respecting rate limits
 * Avoid hitting Telegram API rate limit https://core.telegram.org/bots/faq#broadcasting-to-users
 */
async function editMessageWithCompletionEvery3Seconds(
        ctx: Context,
        messageId: number,
        message: string,
        lastUpdate: number,
        lastSentMessage: string,
        isLastMessage = false,
): Promise<{ timestamp: number; lastMessage: string }> {
        const now = Date.now();
        const has2SecondsPassed = now - lastUpdate >= 2000;
        const displayMessage = message + (isLastMessage ? '' : '...');

        if ((isLastMessage || has2SecondsPassed) && displayMessage !== lastSentMessage) {
                try {
                        await ctx.api.editMessageText(ctx.chat!.id, messageId, toTelegramMarkdown(displayMessage), {
                                parse_mode: 'Markdown',
                        });
                        return { timestamp: now, lastMessage: displayMessage };
                } catch (error) {
                        if (error instanceof Error && error.message.includes('message is not modified')) {
                                return { timestamp: lastUpdate, lastMessage: lastSentMessage };
                        }
                        console.warn(MARKDOWN_ERROR_MESSAGE, displayMessage);
                        try {
                                await ctx.api.editMessageText(ctx.chat!.id, messageId, displayMessage);
                                return { timestamp: now, lastMessage: displayMessage };
                        } catch (fallbackError) {
                                console.error(`Failed to edit message ${messageId} in chat ${ctx.chat!.id}:`, fallbackError);
                                return { timestamp: lastUpdate, lastMessage: lastSentMessage };
                        }
                }
        }

        return { timestamp: lastUpdate, lastMessage: lastSentMessage };
}

1	import { Context } from 'grammy';	1✔
2	import { Action } from 'grammy-auto-chat-action-types';
3	import { Audio, Message, ParseMode, PhotoSize, Voice } from 'grammy-types';
4	import { transcribeAudio } from '@/service/TelegramService.ts';	1✔
5	import { toTelegramMarkdown } from '@/util/MarkdownUtils.ts';	1✔
6
7	const MARKDOWN_ERROR_MESSAGE = 'Error on markdown parse_mode, message:';	1✔
8
9	declare module 'grammy' {
10	interface Context {
11	replyWithQuote(
12	output: string,
13	config?: { parse_mode: ParseMode },
14	): Promise<Message.TextMessage>;
15
16	replyWithVisionNotSupportedByModel(): Promise<Message.TextMessage>;
17
18	startTypingIndicator(): number;
19
20	replyInChunks(output: string): void;
21
22	streamReply(
23	reader: ReadableStreamDefaultReader<Uint8Array>,
24	onComplete: (completedAnswer: string) => Promise<void>,
25	responseMap?: (responseBody: string) => string,
26	lastResult?: string,
27	): Promise<void>;
28
29	extractContextKeys(): Promise<{
30	userId: number;
31	userKey: string;
32	contextMessage?: string;
33	audio?: Voice \| Audio;
34	photos?: PhotoSize[];
35	caption?: string;
36	quote?: string;
37	}>;
38
39	chatAction: Action \| undefined;
40	}
41	}
42
43	/**	1✔
44	* Reply to a message with quoting the original message
45	*/	×
46	Context.prototype.replyWithQuote = function (	×
47	this: Context,
48	output: string,	×
49	config?: { parse_mode: ParseMode },	×
50	) {
51	return this.reply(output, {	×
52	reply_to_message_id: this.message?.message_id,	×
53	...config,	×
54	});	×
55	};	×
56
57	/**	1✔
58	* Reply that the model doesn't support vision capabilities
59	*/	1✔
60	Context.prototype.replyWithVisionNotSupportedByModel = function (	1✔
61	this: Context,
62	) {
63	return this.replyWithQuote('esse modelo não suporta leitura de foto');	2✔
64	};	1✔
65
66	/**	1✔
67	* Start typing indicator that persists by re-sending every 4 seconds
68	*/	1✔
69	Context.prototype.startTypingIndicator = function (this: Context): number {	1✔
70	this.chatAction = 'typing';	2✔
71	return setInterval(() => {	×
72	this.chatAction = 'typing';	×
73	}, 4000);	×
74	};	1✔
75
76	/**	1✔
77	* Split a large response into multiple message chunks
78	*/	1✔
79	Context.prototype.replyInChunks = function (	1✔
80	this: Context,
81	output: string,	1✔
82	): void {
83	if (output.length > 4096) {	3✔
84	const outputChunks = output.match(/[\s\S]{1,4093}/g)!;	4✔
85
86	outputChunks.forEach((chunk, index) => {	4✔
87	const isLastChunk = index === outputChunks.length - 1;	6✔
88	const chunkOutput = `${chunk}${isLastChunk ? '' : '...'}`;	6✔
89	const sanitizedOutput = toTelegramMarkdown(chunkOutput);	6✔
90
91	this.replyWithQuote(sanitizedOutput, { parse_mode: 'Markdown' })	12✔
92	.catch(() => {	×
93	console.warn(MARKDOWN_ERROR_MESSAGE, chunkOutput);	×
94	this.replyWithQuote(chunkOutput);	×
95	});	×
96	});	4✔
97	return;	4✔
98	}	4✔
99
100	const sanitizedOutput = toTelegramMarkdown(output);	4✔
101	this.replyWithQuote(sanitizedOutput, { parse_mode: 'Markdown' })	12✔
102	.catch(() => {	4✔
103	console.warn(MARKDOWN_ERROR_MESSAGE, output);	5✔
104	this.replyWithQuote(output);	5✔
105	});	4✔
106	};	1✔
107
108	/**	1✔
109	* Stream a response to the user with periodic updates
110	*/	1✔
111	Context.prototype.streamReply = async function (	1✔
112	this: Context,
113	reader: ReadableStreamDefaultReader<Uint8Array>,	1✔
114	onComplete: (completedAnswer: string) => Promise<void>,	1✔
115	responseMap?: (responseBody: string) => string,	1✔
116	lastResult?: string,	1✔
117	): Promise<void> {
118	const { message_id } = await this.replyWithQuote('processando...');	2✔
119	let result = lastResult \|\| '';	2✔
120	let lastUpdate = Date.now();	2✔
121	let lastSentMessage = '';	2✔
122
123	while (true) {	2✔
124	const { done, value } = await reader.read();	5✔
125	if (done) break;	5✔
126
127	const chunk = decodeStreamResponseText(value, responseMap);	7✔
128	result += chunk;	7✔
129
130	if (result.length > 4093) {	×
131	result = result.removeThinkingChatCompletion()	×
132	.convertBlackBoxWebSearchSourcesToMarkdown();	×
133
134	if (result.length > 4093) {	×
135	const remainingChunk = result.substring(4093) + chunk;	×
136	result = result.substring(0, 4093);	×
137
138	const updateResult = await editMessageWithCompletionEvery3Seconds(	×
139	this,	×
140	message_id,	×
141	result,	×
142	lastUpdate,	×
143	lastSentMessage,	×
144	true,	×
145	);
146	lastUpdate = updateResult.timestamp;	×
147	lastSentMessage = updateResult.lastMessage;	×
148	onComplete(result);	×
149	return this.streamReply(	×
150	reader,	×
151	onComplete,	×
152	responseMap,	×
153	remainingChunk,	×
154	);
155	}	×
156	}	✔
157
158	const updateResult = await editMessageWithCompletionEvery3Seconds(	7✔
159	this,	7✔
160	message_id,	7✔
161	result,	7✔
162	lastUpdate,	7✔
163	lastSentMessage,	7✔
164	);
165	lastUpdate = updateResult.timestamp;	7✔
166	lastSentMessage = updateResult.lastMessage;	7✔
167	}	7✔
168
169	let sanitizedResult = result.removeThinkingChatCompletion()	2✔
170	.convertBlackBoxWebSearchSourcesToMarkdown();	2✔
171
172	if (sanitizedResult.length > 4093) {	×
173	const remainingChunk = sanitizedResult.substring(4093);	×
174	sanitizedResult = sanitizedResult.substring(0, 4093) + '...';	×
175	this.replyInChunks(remainingChunk);	×
176	}	×
177
178	this.api.editMessageText(this.chat!.id, message_id, toTelegramMarkdown(sanitizedResult), {	2✔
179	parse_mode: 'Markdown',	2✔
180	})	×
181	.catch(() => {	×
182	console.warn(MARKDOWN_ERROR_MESSAGE, sanitizedResult);	×
183	this.api.editMessageText(this.chat!.id, message_id, sanitizedResult);	×
184	});	×
185
186	onComplete(result);	2✔
187	};	1✔
188
189	/**	1✔
190	* Extract common context keys from the message
191	*/	1✔
192	Context.prototype.extractContextKeys = async function (this: Context) {	1✔
193	const userId = this.from?.id!;	2✔
194	const userKey = `user:${userId}`;	2✔
195	const audio = this.message?.voice \|\| this.message?.audio;	2✔
196	const contextMessage = await getTextMessage(userId, userKey, this, audio);	2✔
197	const photos = this.message?.photo;	2✔
198	const caption = this.message?.caption;	2✔
199	const quote = this.message?.reply_to_message?.text;	×
200
201	return { userId, userKey, contextMessage, audio, photos, caption, quote };	18✔
202	};	1✔
203
204	/**	1✔
205	* Helper function to get text from a message, transcribing audio if needed
206	*/	1✔
207	function getTextMessage(	1✔
208	userId: number,	1✔
209	userKey: string,	1✔
210	ctx: Context,	1✔
211	audio?: Voice,	1✔
212	): Promise<string \| undefined> {
213	if (audio) {	×
214	return transcribeAudio(userId, userKey, ctx, audio);	×
215	}	×
216	return Promise.resolve(ctx.message?.text);	2✔
217	}	2✔
218
219	/**	1✔
220	* Decode text from the response stream
221	*/	1✔
222	function decodeStreamResponseText(	1✔
223	responseMessage: Uint8Array,	1✔
224	responseMap?: (responseBody: string) => string,	1✔
225	): string {
226	const decoder = new TextDecoder();	3✔
227	const decodedText = decoder.decode(responseMessage);	3✔
228	return responseMap ? responseMap(decodedText) : decodedText;	×
229	}	3✔
230
231	/**	1✔
232	* Edit a message with updated content, respecting rate limits
233	* Avoid hitting Telegram API rate limit https://core.telegram.org/bots/faq#broadcasting-to-users
234	*/	1✔
235	async function editMessageWithCompletionEvery3Seconds(	1✔
236	ctx: Context,	1✔
237	messageId: number,	1✔
238	message: string,	1✔
239	lastUpdate: number,	1✔
240	lastSentMessage: string,	1✔
241	isLastMessage = false,	1✔
242	): Promise<{ timestamp: number; lastMessage: string }> {
243	const now = Date.now();	3✔
244	const has2SecondsPassed = now - lastUpdate >= 2000;	3✔
245	const displayMessage = message + (isLastMessage ? '' : '...');	×
246
247	if ((isLastMessage \|\| has2SecondsPassed) && displayMessage !== lastSentMessage) {	×
248	try {	×
NEW 249	await ctx.api.editMessageText(ctx.chat!.id, messageId, toTelegramMarkdown(displayMessage), {	×
250	parse_mode: 'Markdown',	×
251	});	×
252	return { timestamp: now, lastMessage: displayMessage };	×
253	} catch (error) {	×
254	if (error instanceof Error && error.message.includes('message is not modified')) {	×
255	return { timestamp: lastUpdate, lastMessage: lastSentMessage };	×
256	}	×
257	console.warn(MARKDOWN_ERROR_MESSAGE, displayMessage);	×
258	try {	×
259	await ctx.api.editMessageText(ctx.chat!.id, messageId, displayMessage);	×
260	return { timestamp: now, lastMessage: displayMessage };	×
261	} catch (fallbackError) {	×
262	console.error(`Failed to edit message ${messageId} in chat ${ctx.chat!.id}:`, fallbackError);	×
263	return { timestamp: lastUpdate, lastMessage: lastSentMessage };	×
264	}	×
265	}	×
266	}	×
267
268	return { timestamp: lastUpdate, lastMessage: lastSentMessage };	12✔
269	}	3✔

lucasliet / llm-telegram-bot / 21789375408

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous