Session Bridge for OpenClaw

Cross-channel session context bridging for OpenClaw. Load conversation context from any session into any channel — webchat, Telegram, Discord, WhatsApp, etc.

Supports Gemini, OpenAI, and OpenRouter as LLM providers for naming and summarization.

What It Does

When you're working on a project in webchat and want to continue on Telegram (or vice versa), Session Bridge lets you:

1. List all your sessions across channels with auto-generated names
2. Load a session's full context into your current chat via LLM summarization
3. Name sessions with custom titles for easy reference
4. Browse historical/archived sessions

How It Works

/session list              → See all active sessions with names
/session list --all        → Include historical/archived sessions
/session load <# or name>  → Load session context into current chat
/session name <title>      → Name the current session
/session rename <# or name> <title> → Rename any session
/session help              → Show commands + active provider

Example

You (on Telegram):  /session list

📋 Active Sessions (via gemini/gemini-3-pro-preview):
1. 💬 session "Electron Window Manager Build", webchat, 3m ago
2. ⏰ cron "Proton Mail Bridge Sync", 1d ago (52 runs)
3. 💬 session "Side Project", telegram, 2h ago

You (on Telegram):  /session load 1

✅ Session "Electron Window Manager Build" loaded via gemini!
Context has been injected into your current session...

Now the AI assistant in your Telegram chat has full context from your webchat conversation and can continue seamlessly.

How Context Transfer Works

1. Reads the session's JSONL transcript
2. Sends it to your configured LLM provider for extensive summarization
3. Summary is formatted as structured JSON with: topic, decisions, technical details, current status, open items, preferences, key files, and people
4. Summary is injected into the current session via OpenClaw's system event mechanism
5. The AI assistant sees the summary and can continue the work naturally

Batched Session Naming

When listing sessions, all unnamed sessions are named in a single batched LLM call — the plugin sends all conversation snippets in one prompt and gets all titles back at once. This is faster and avoids rate limits compared to individual API calls per session.

Installation

From source

git clone https://github.com/clockworksquirrel/session-bridge-openclaw.git
cp -r session-bridge-openclaw ~/.openclaw/extensions/session-bridge

Manual

Copy the files to ~/.openclaw/extensions/session-bridge/:

~/.openclaw/extensions/session-bridge/
├── index.ts
├── openclaw.plugin.json
└── package.json

Enable in config

Add to your ~/.openclaw/openclaw.json:

{
  "plugins": {
    "entries": {
      "session-bridge": {
        "enabled": true,
        "config": {
          "provider": "gemini",
          "geminiModel": "gemini-3-pro-preview",
          "maxTranscriptChars": 200000
        }
      }
    }
  }
}

Then restart the gateway:

openclaw gateway restart

LLM Providers

Session Bridge supports three providers. Set provider in your config to switch.

Gemini (default)

{
  "provider": "gemini",
  "geminiModel": "gemini-3-pro-preview",
  "geminiApiKey": "your-key-here"
}

API key resolution: plugin config → GEMINI_API_KEY env → macOS Keychain (service: gemini, account: api-key)

OpenAI

{
  "provider": "openai",
  "openaiModel": "gpt-4o",
  "openaiApiKey": "sk-..."
}

API key resolution: plugin config → OPENAI_API_KEY env → macOS Keychain (service: openai, account: api-key)

The openaiBaseUrl option supports any OpenAI-compatible API (Azure, local LLMs, etc.):

{
  "provider": "openai",
  "openaiModel": "llama-3.3-70b",
  "openaiApiKey": "your-key",
  "openaiBaseUrl": "http://localhost:11434/v1"
}

OpenRouter

{
  "provider": "openrouter",
  "openrouterModel": "anthropic/claude-sonnet-4",
  "openrouterApiKey": "sk-or-..."
}

API key resolution: plugin config → OPENROUTER_API_KEY env → macOS Keychain (service: openrouter, account: api-key)

Any model on OpenRouter's model list works.

Configuration

Key	Default	Description
provider	gemini	LLM provider: gemini, openai, or openrouter
geminiModel	gemini-3-pro-preview	Gemini model for summarization and naming
geminiApiKey	(auto-detected)	Gemini API key
openaiModel	gpt-4o	OpenAI model
openaiApiKey	(auto-detected)	OpenAI API key
openaiBaseUrl	https://api.openai.com/v1	Custom OpenAI-compatible base URL
openrouterModel	anthropic/claude-sonnet-4	OpenRouter model
openrouterApiKey	(auto-detected)	OpenRouter API key
maxTranscriptChars	200000	Max transcript characters sent for summarization

Requirements

• OpenClaw (2026.2.x or later)
• An API key for at least one supported provider

Session Types

The list view shows emoji tags for different session types:

Emoji	Type	Description
💬	session	Regular chat session (DM, webchat)
💬	dm	Direct message session
⏰	cron	Cron job session (collapsed with run count)
🪝	hook	Webhook session
🤖	subagent	Sub-agent session
👥	group	Group chat session
📢	channel	Channel/room session
🧵	thread/topic	Thread or topic session
⚡	slash	Slash command session
📜	archived	Historical session (--all flag)

Changelog

2026-02-11 (v1.1.0)

• Multi-provider support: Added OpenAI and OpenRouter alongside Gemini. Any OpenAI-compatible API works via openaiBaseUrl.
• Batched session naming: All unnamed sessions are now named in a single LLM call instead of one-by-one. Fixes the issue where only 1-2 sessions got named before rate limits kicked in.
• Provider display: /session list and /session help show which provider and model are active.

2026-02-11 (v1.0.0)

• Initial release with Gemini support, /session list/load/name/rename/help commands.

License

MIT

Credits

Built by Vera 🌀 for Josh.