OmniBridge

OmniBridge is an AI-native binary protocol gateway: it parses known protocols at high speed and learns unknown ones automatically using an LLM, then persists the learned parser for future traffic.

If your data streams are evolving faster than hand-written decoders, OmniBridge gives you a practical way to keep up.

---

• ⚡ Fast path first: Known signatures route directly to existing parsers via a Trie-based dispatcher.
• 🧠 AI discovery mode: Unknown packets trigger LLM-assisted parser generation with Zero-Config Signature Detection.
• 🔁 Self-healing parsers: If a learned parser fails at runtime, OmniBridge attempts automatic repair by consulting the LLM with the error context.
• 💾 Persistent learning: Generated parsers are cached in-memory and saved in ./storage for persistence.
• 🔌 Provider flexibility: Works with Gemini (cloud) and Ollama (local).
• 🧪 Execution Safety: Dynamic parsers run with 50ms timeout protection and panic recovery to ensure system stability.

---

1. Ingest: Raw bytes are received from simulation or TCP server modes.
2. Dispatch: Using a Trie (Prefix Tree), the system performs a longest-prefix match on the incoming data to identify the protocol.
3. Parse: If a parser exists, it is executed within the Dynamic Engine.
4. Discover: On a cache miss, the AI identifies the signature and generates a Go Parse function.
5. Cache + Persist: The new parser is compiled (via yaegi), cached for performance, and saved to disk.
6. Repair: If a parser throws a runtime error, the Self-Healing Loop triggers an AI-assisted fix.

---

🚀 Quick Start

1) Prerequisites

• Go 1.25+
• One LLM provider:
  • Gemini: set GEMINI_API_KEY
  • Ollama: local Ollama server running

2) Install

git clone https://github.com/chuanjin/OmniBridge.git
cd OmniBridge
go mod tidy

3) Configure environment

Create a .env file:

# Needed only for Gemini provider
GEMINI_API_KEY=your_api_key_here

4) Run in simulation mode (default)

go run cmd/server/main.go --provider gemini --model gemini-2.0-flash

Run with Ollama:

go run cmd/server/main.go --provider ollama --model deepseek-coder:1.3b

5) Run as TCP gateway

go run cmd/server/main.go --mode server --addr :8080 --provider gemini --model gemini-2.0-flash

Send binary data to it from your client; OmniBridge will parse known signatures and discover unknown ones.

---

🐳 Docker

Build and run:

docker build -t omnibridge .
docker run --rm -p 8080:8080 --env GEMINI_API_KEY=$GEMINI_API_KEY omnibridge

---

📁 Project layout

• cmd/server/ — CLI entrypoint (simulation + TCP server modes)
• internal/parser/ — dispatcher, discovery service, parser manager, dynamic engine
• internal/logger/ — structured logging setup
• agents/ — system prompt(s) used for parser generation
• seeds/ — built-in parser seeds loaded at startup
• examples/ — sample protocol data
• storage/ — learned parsers + manifest (created at runtime)

---

🛠️ Internal Mechanics

Trie Dispatcher

OmniBridge uses a Prefix Tree (Trie) to manage protocol signatures. This enables efficient routing even with variable-length signatures, ensuring the longest match is always prioritized.

Dynamic Engine & Caching

Parsers are implemented as Go code generated by AI. To ensure high performance:

• JIT Compilation: Code is compiled at runtime using the yaegi interpreter.
• Concurrent Caching: Compiled functions are cached in a thread-safe map, avoiding redundant compilation overhead for future packets.

Execution Safety

Running AI-generated code requires guardrails. OmniBridge provides:

• Timeout Protection: Every parser execution is capped at 50ms.
• Panic Recovery: The system traps runtime panics (e.g., out-of-bounds access) and routes them to the repair cycle.
• Restricted Stdlib: Parsers only have access to safe packages like encoding/binary, math, and bytes.

---

🧭 Project Roadmap

OmniBridge is actively evolving. High-priority areas include:

• Enhanced Observability: Exporting metrics via Prometheus/OpenTelemetry.
• Hardened Validation: Pre-execution static analysis of AI-generated code.
• Stateful Protocols: Support for protocols requiring sequence tracking or multi-packet assembly.

---

🧪 MCP Integration

OmniBridge can run as an MCP Server (Model Context Protocol), exposing its protocol discovery and parsing capabilities to AI applications and agents.

Running as MCP Server

go run cmd/server/main.go --mode mcp --provider gemini --model gemini-2.0-flash

Available Resources

• protocol://list - List all known protocols with signatures
• protocol://manifest - Complete manifest mapping

Available Tools

• parse_binary - Parse hex-encoded binary data
• discover_protocol - Trigger AI-based protocol discovery
• list_protocols - List all available protocols

Available Prompts

• protocol_discovery - Template for discovering new protocols
• parser_repair - Template for fixing broken parsers

Example MCP Client Configuration (Claude Desktop)

{
  "mcpServers": {
    "omnibridge": {
      "command": "go",
      "args": ["run", "/path/to/OmniBridge/cmd/server/main.go", "--mode", "mcp", "--provider", "gemini"]
    }
  }
}

---

🤝 Contributing

Issues and PRs are welcome. If you have a target protocol family (CAN, telemetry, industrial buses, custom IoT frames), open an issue with sample payloads and expected fields.

---

📜 License

MIT — see LICENSE.