WebSocket and streaming

Tenet provides two different WebSocket protocols. Choose the endpoint by client type; they are not wire-compatible.

Endpoint	Protocol	Intended clients
ws://127.0.0.1:10444/ws/<instance-id>	JSON-RPC/MCP text frames	IDAPython and local interactive integrations
ws://127.0.0.1:<port> from --ws-port	Binary FlatBuffers	Tauri Web frontend and high-throughput visualization

Hub JSON WebSocket

Start Tenet in Hub mode. When a trace is opened in the Tauri desktop application, its sidecar starts in Hub mode or joins the existing Hub:

./tenet trace.bin --mcp-hub

Discover instance IDs over http://127.0.0.1:10444/mcp, then connect to:

ws://127.0.0.1:10444/ws/<instance-id>

The URL binds the connection to one trace. Text frames carry the same JSON-RPC/MCP requests and responses used by tools/call. The endpoint supports ping/pong and close handling, validates RFC 6455 framing and UTF-8, rejects binary application messages, and limits payload size.

The Hub JSON endpoint is deliberately loopback-only. It is designed for local tools such as the IDA Pro bridge, not for direct exposure to a network. If the selected Worker exits, the Hub closes sessions bound to that instance so clients cannot continue querying stale state.

FlatBuffers WebSocket

Start the independent binary service on an explicit port:

./tenet trace.bin --ws-port 9090

The endpoint is ws://127.0.0.1:9090. The current service always binds to the local loopback interface; no CLI option can change the listen host.

Windows limitation

On Windows, TENET_ENABLE_WS defaults to off and cannot currently be enabled: the uSockets event loop has POSIX epoll/kqueue backends but no Windows IOCP backend wired into this build. Consequently, --ws-port and the FlatBuffers WebSocket channel used by the Tauri Web frontend are not yet available on Windows. MCP Hub’s JSON WebSocket does not use this uSockets/FlatBuffers service and is unaffected; CLI batch analysis, TUI, TCP JSON-RPC, and MCP also remain available.

Protocol

• Transport: WebSocket on the configured port.
• Schema: FlatBuffers (.fbs definitions in the Tenet source).
• Encoding: binary frames with no JSON parsing overhead.
• Use case: high-throughput frontend queries, analysis jobs, and visualization.

This service is separate from MCP Hub. Do not connect the IDAPython bridge or a JSON-RPC client to --ws-port, and do not send FlatBuffers messages to port 10444.

Streaming index

WebSocket transport is also distinct from --streaming. Streaming mode tails a growing trace and incrementally updates its index:

./tenet trace.bin --streaming

Use it to pre-build index data while capture is still active. It does not itself create either WebSocket endpoint.

Operational boundaries

• Both WebSocket services require an active Tenet process with an open trace.
• Hub JSON sessions are instance-scoped and close when their Worker disappears.
• FlatBuffers schema stability is not guaranteed across versions; update custom clients with Tenet.
• Bound high-volume requests and debounce cursor updates in interactive clients.