v4.0.0 — Streaming Preview + Effort Control + Export

telegram-ai-bridge v4.0.0

45 commits since v3.0.0. Major usability and reliability upgrade.

New Features

• Streaming Preview — see Claude's thinking in real-time as messages stream in
• /effort Command — per-chat thinking depth control (low/medium/high/max), backend-aware
• /export Command — dump War Room conversation as Markdown file
• /cancel Command — abort running tasks mid-execution
• /help Command — TG slash command menu
• Output Relay — images, files, and long code blocks relayed back to Telegram
• Message Reply Context — reply to specific messages for contextual conversations
• Smart Text Splitting — long non-code text → summary preview + .md file attachment
• Session Resume — resume any session directly (not just peek)
• Session List Enhancement — topic-first display, noise filtering, transparency notes

Reliability

• Send Retry — automatic retry on failed message sends
• File Ref Protection — prevent file reference loss during message splitting
• Graceful Shutdown — clean up orphaned progress messages on shutdown
• Bridge Hint Injection — prevent CC from calling TG API directly

Fixes

• Fix grammy multipart proxy issue for photo/file upload
• Fix /cancel abort handling
• Fix /status Markdown parse failure crash
• Expand sendable file extensions (txt, md, json, py, sh, zip, etc.)

Full changelog: v3.0.0...v4.0.0

v3.0.0 — Full Claude Code, Everywhere

What's New

The biggest release yet. telegram-ai-bridge is no longer just a bridge — it's a multi-instance Claude Code command center you run from your phone.

Multi-Instance Parallel Sessions

Run N parallel Claude Code instances, each with its own Telegram bot. All instances share memory (~/.claude/, RecallNest, MCP servers) automatically — what you tell one, the others already know. Per-bot personas via workspace CLAUDE.md.

War Room — Multi-CC Command Center

Put 4 CC bots in one group. @mention to dispatch — each bot reads others' replies via shared context (Redis-backed). Two modes: A2A (auto-debate) and War Room (@mention only).

Bidirectional Media Relay

CC output now flows back to Telegram:

• Screenshots: SDK tool results (peekaboo, take_screenshot) → auto-sent as photos
• Files: CC creates/references files → bridge detects paths and sends as attachments
• Long code: >4000 chars with >60% code → file attachment + preview summary
• Long text: >4000 chars non-code → preview + response.md attachment

War Room Export

/export dumps the entire cross-bot conversation as a timestamped Markdown file — full audit trail of every bot's contribution in War Room sessions.

Smarter Session List

Session picker redesigned for readability:

• Topic first: user's first message is the primary label, not metadata
• Noise filtered: bridge hints ([系统提示:...]) and <local-command-> stripped
• Cleaner layout: removed redundant CLI/SDK source tags and home-dir project names
• Unified view: lists both terminal and TG sessions — resume terminal sessions from your phone

New Commands

Command	Description
/help	Categorized command reference
/cancel	Abort running task from phone
/dir	Switch working directory
/cron	Manage scheduled tasks
/doctor	Health check

Interaction Enhancements

• Message reply context: Reply to any TG message → quoted text included as context
• Bridge hint injection: CC knows to output file paths for auto-relay, won't try to call TG API directly
• Smart file extension support: png, jpg, pdf, docx, xlsx, csv, html, svg, txt, md, json, js, ts, py, sh, yaml, zip, gz, and more
• Transparent bridge: TG bot inherits all local CC capabilities (skills, MCP servers, hooks) — no bridge-level reimplementation needed

Security & Stability

• A2A bus bound to 127.0.0.1 (was 0.0.0.0)
• A2A group-chat-only enforcement (prevents DM cross-leak)
• SDK updated: @anthropic-ai/claude-agent-sdk 0.2.63 → 0.2.81

Breaking Changes

• Version bump from 2.x to 3.0 — A2A was temporarily removed then restored with group-only restriction
• /relay command added then architecture simplified — A2A broadcast replaces point-to-point relay

---

Full changelog: v2.2.0...v3.0.0

v2.2.0 — Pluggable Shared-Context Backend

What's New

Shared-context storage is now pluggable — choose the backend that fits your deployment:

Backend	Dependencies	Concurrency	Best For
sqlite (default)	None (built-in)	WAL mode, single-writer	Single bot, low concurrency
json	None (built-in)	Atomic write (tmp+rename)	Zero-dependency deployment
redis	ioredis	Native concurrency + TTL	Multi-bot, Docker environment

Highlights

• Factory pattern with per-backend modules in shared-context/
• Redis backend eliminates "database is locked" on concurrent writes
• JSON backend for zero-dependency deployments
• A2A handler now reads shared context history before responding
• Backward compatible — defaults to SQLite

Configuration

Set sharedContextBackend in config.json: sqlite | json | redis

Full Changelog: v2.1.0...v2.2.0