Codex Telegram Coding Assistant
https://github.com/coleam00/codex-telegram-coding-assistant- Category
- AI Agents
- Rank
- No. 1610Tools index
Previous survey · No. 1616 ·
- Pricing
- Open Source
- Type
- AGENT
- Builder
- coleam00
- GitHub
- 59 stars
- Date
About
Codex-powered coding assistant that integrates with Telegram so you can send coding requests from your phone.
What it does
It maintains separate Codex conversations for each user and project directory, streams replies through Telegram, and persists session state across restarts. For frontend requests, it can add browser verification instructions, capture screenshots, and expose Browserbase session replays when the optional Stagehand setup is available.
Why it's ranked here
The project combines remote code execution, project-specific conversation context, and visible frontend checks in one deployable bot. Its strongest idea is operational continuity: changing projects starts fresh context while preserving prior threads. The default unattended write permissions make it capable, but also demand a carefully isolated workspace and trusted users.
What's good
Sessions are isolated per Telegram user and saved as readable JSON. Each working directory receives its own Codex thread, reducing context leakage between projects. Responses stream as work progresses. A reset clears conversation history without losing the selected project. Optional browser automation can navigate interfaces, interact with controls, capture evidence, and produce replay links.
Tradeoffs
Setup requires Docker, Telegram credentials, four Codex authentication values, and a GitHub token according to the documentation. Browser verification adds three more credentials and an external Browserbase account. Codex runs with workspace writes and no approval prompts by default. Sessions use local JSON files, and the supplied code shows no explicit Telegram user allowlist. Frontend detection relies on keyword matching, so unrelated mentions can trigger verification instructions.
How to use it well
Use it for trusted, remote maintenance tasks where a developer wants to direct Codex from Telegram and inspect streamed progress. Mount only intended projects into the workspace, restrict access to the bot, and enable Browserbase when visual proof matters. It complements an editor and deployment pipeline; it does not provide repository hosting, production isolation, or a full graphical development environment.
Technical notes+
src/index.ts builds a Telegraf bot and registers command plus text handlers. src/codex/client.ts wraps @openai/codex-sdk, starts or resumes threads with skipGitRepoCheck: true, and streams turns. src/session/manager.ts stores one JSON record per user under telegram_sessions, clears thread_id when the working directory changes, and defaults to /workspace. src/codex/prompt-builder.ts detects frontend work through substring keywords and appends Stagehand screenshot instructions. src/config/env.ts validates Telegram and Codex credentials but does not validate the GH_TOKEN that README.md calls required. package.json defines the validate script twice, so the later entry replaces the earlier one in standard JSON parsing.
Observed
- License
- MIT
- Primary language
- TypeScript using ECMAScript modules
- Install and deployment
- Node package with Docker and Docker Compose deployment documented
- User interface
- Telegram bot built with Telegraf
- Core integration
- OpenAI Codex SDK with streamed thread execution
- Optional interface
- Stagehand MCP through Browserbase for browser automation and session replays
- Persistence
- Per-user sessions stored as local JSON files, with Codex session data mounted separately
- Repository structure
- No test directory appears in the supplied project tree; package scripts reference standalone test files
Read from README.md, package.json, src/index.ts, src/config/env.ts, src/types/index.ts, src/codex/types.ts, src/codex/client.ts, src/codex/events.ts, src/session/types.ts, src/session/manager.ts, src/codex/prompt-builder.ts, src/bot/commands/help.ts, src/bot/commands/reset.ts, src/bot/commands/start.ts, src/bot/commands/getcwd.ts.
What it can do
Generate code from natural language prompts
Text description of coding task → Code snippets or programs
Debug existing code
Code with errors or bugs → Corrected code with fixes
Explain code functionality
Code snippets → Plain language explanations
Convert code between programming languages
Code in one programming language → Equivalent code in another language
Optimize code performance
Existing code → Improved, more efficient code
Review code for best practices
Code snippets → Code review feedback and suggestions
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.