Architecture

IM Platform (Telegram / Discord / Feishu)
        |
        | InboundMessage
        v
   +-----------+     +------------------+
   |  Adapter   |---->| Bridge Manager   |  (orchestrator)
   +-----------+     |  |- Channel Router     -> session binding
                     |  |- Conversation Engine -> LLM streaming
                     |  |- Permission Broker   -> tool approval flow
                     |  |- Delivery Layer      -> chunking, retry, dedup
                     +------------------+
                            |
                            | Host Interfaces (DI)
                            v
                     +------------------+
                     | Host Application |  (implements BridgeStore,
                     |                  |   LLMProvider, etc.)
                     +------------------+

All bridge modules access host services through a DI context (getBridgeContext()), never through direct imports. This means you can plug the bridge into any Node.js application by implementing four interfaces.

Quick Start

1. Install

npm install claude-to-im

Or clone this repo and install dependencies:

git clone https://github.com/op7418/Claude-to-IM.git
cd Claude-to-IM
npm install

2. Implement Host Interfaces

The bridge requires four interfaces. See docs/development.md for the full specification of each interface.

import { initBridgeContext } from 'claude-to-im/context';
import type { BridgeStore, LLMProvider, PermissionGateway, LifecycleHooks } from 'claude-to-im/host';

const store: BridgeStore = { /* your persistence layer (~30 methods) */ };
const llm: LLMProvider = { /* wraps Claude Code SDK streamChat */ };
const permissions: PermissionGateway = { /* resolves pending tool permissions */ };
const lifecycle: LifecycleHooks = { /* optional start/stop callbacks */ };

initBridgeContext({ store, llm, permissions, lifecycle });

3. Start the Bridge

import * as bridgeManager from 'claude-to-im/bridge-manager';

await bridgeManager.start();

const status = bridgeManager.getStatus();
// { running: true, adapters: [{ channelType: 'telegram', running: true, ... }] }

4. Run the Example

A self-contained example with in-memory store and echo LLM is included:

npx tsx src/lib/bridge/examples/mock-host.ts

Configuration

All settings are read through BridgeStore.getSetting(key). Your host application decides how to store and surface these values (database, env vars, config file, UI settings panel, etc.).

Required Settings

| Key | Description | |-----|-------------| | remote_bridge_enabled | Master switch — "true" to enable the bridge | | bridge_{adapter}_bot_token | Bot token for the platform (e.g. bridge_telegram_bot_token) | | bridge_{adapter}_allowed_users | Comma-separated user IDs authorized to use the bridge |

Optional Settings

| Key | Description | Default | |-----|-------------|---------| | bridge_auto_start | Auto-start bridge on app launch | "false" | | bridge_{adapter}_enabled | Per-adapter toggle | "false" | | bridge_{adapter}_stream_enabled | Enable streaming previews | "true" | | bridge_default_cwd | Default working directory for new sessions | $HOME | | bridge_model | Default Claude model | Host decides |

Replace {adapter} with telegram, discord, or feishu.

Limitations

Before adopting this library, be aware of the following constraints:

You Must Implement the Host Interfaces

This is a library, not a standalone application. You need to provide:

• BridgeStore — A persistence layer with ~30 methods covering settings, sessions, messages, channel bindings, audit logs, dedup tracking, permission links, and channel offsets. This is the largest integration surface. See the full interface definition in src/lib/bridge/host.ts.
• LLMProvider — A wrapper around your LLM client that returns a ReadableStream<string> of SSE-formatted events. The stream format must match the Claude Code SDK's event protocol (text, tool_use, tool_result, permission_request, status, result events). See docs/development.md for the full event format spec.
• PermissionGateway — A way to resolve pending tool permissions from the Claude Code SDK.
• LifecycleHooks — Optional callbacks for bridge start/stop events.

LLM Stream Format

The LLMProvider.streamChat() must return SSE-formatted strings matching the Claude Code SDK's event protocol. If you are not using the Claude Code SDK, you need to adapt your LLM client's output to match this format. This is not a generic "chat completion" interface.

No Built-in Persistence

The bridge does not bundle any database driver. You provide all persistence through BridgeStore. This gives you full control but means you need to implement storage for sessions, messages, bindings, audit logs, dedup keys, permission links, and channel offsets.

Session Locking

The bridge uses a session lock mechanism (acquireSessionLock / renewSessionLock / releaseSessionLock) to serialize messages within the same session. Your BridgeStore implementation must provide atomic lock operations. For single-process deployments, in-memory locks work fine. For multi-process deployments, you need distributed locking (e.g., database-backed).

Platform Bot Setup

You still need to create bots on each platform and obtain tokens:

• Telegram: Create a bot via @BotFather
• Discord: Create an application in the Developer Portal with Message Content Intent enabled
• Feishu: Create an app in the Developer Console with IM permissions

Documentation

| Document | Description | |----------|-------------| | Development Guide | Host interface specs, SSE format, adapter development, step-by-step integration tutorial | | Architecture | Module dependency graph, message flows, design decisions | | Security | Threat model, mitigations, deployment recommendations | | Contributing | Dev setup, code style, testing guide | | Migration | Before/after import patterns for migrating from direct imports |

Project Structure

src/
  lib/bridge/
    context.ts              # DI container (initBridgeContext / getBridgeContext)
    host.ts                 # Host interface definitions (BridgeStore, LLMProvider, etc.)
    types.ts                # Shared type definitions (messages, bindings, status)
    bridge-manager.ts       # Orchestrator — start/stop, message dispatch, session locks
    channel-adapter.ts      # Abstract base class + adapter registry
    channel-router.ts       # ChannelAddress -> ChannelBinding resolution
    conversation-engine.ts  # LLM stream processing, SSE consumption
    delivery-layer.ts       # Reliable outbound delivery (chunk, retry, dedup, audit)
    permission-broker.ts    # Tool permission forwarding and callback handling
    adapters/
      telegram-adapter.ts   # Telegram Bot API long polling
      discord-adapter.ts    # Discord.js Gateway WebSocket
      feishu-adapter.ts     # Feishu/Lark WSClient
      telegram-media.ts     # Telegram file download/attachment handling
      telegram-utils.ts     # Telegram API helpers
      index.ts              # Side-effect imports for adapter self-registration
    markdown/
      ir.ts                 # Intermediate representation for Markdown AST
      render.ts             # Generic IR -> platform string renderer
      telegram.ts           # Markdown -> Telegram HTML
      discord.ts            # Markdo