Tunnlo connects real-time data to LLM-powered agents that monitor, triage, and act — with smart filtering, multi-agent orchestration, and per-agent dashboards so only what matters hits the right LLM.
See how engineering teams use Tunnlo to turn noisy data streams into automated, intelligent workflows.
Your on-call engineer is drowning in alerts. Tunnlo filters log noise, triages errors with AI, and escalates only what matters — before your team burns out.
Read case study →Your SIEM fires 10,000 alerts a day. 95% are false positives. Tunnlo captures network traffic, classifies threats with AI, and surfaces real attacks in real time.
Read case study →Your Kafka topics are full of user events nobody reads. Tunnlo samples high-volume streams, detects churn signals and sentiment shifts, and alerts your team in real time.
Read case study →Your Google Docs are full of comments and suggestions nobody triages. Tunnlo watches docs in real time, summarizes feedback, evaluates suggested edits, and alerts your team.
Read case study →Route different data sources to specialized LLM agents, compare models side by side, and correlate insights across your entire stack with a single pipeline config.
Read case study →From data ingestion to agent response, Tunnlo handles the entire flow with a modular, extensible architecture.
Ingest from Kafka topics, Google Docs, network traffic, log files, stdin, MCP servers, or build your own with the Adapter SDK. Hot-swappable at runtime.
Rate limiting, deduplication, content matching, adaptive sampling, windowed aggregation, and priority routing built in.
Run multiple agents in parallel — compare LLMs side by side or route different sources to specialized agents. Supports Anthropic, OpenAI, Ollama, LangGraph, CrewAI, and OpenClaw.
Agents respond with structured actions: webhooks, MCP tool calls, or approval-gated operations with human-in-the-loop.
Built-in web dashboard with per-agent response panels, side-by-side comparison view, live metrics, filter drop rates, and token usage tracking.
Self-hosted, no telemetry, no cloud dependency. Your data stays on your infrastructure. Runs fully offline with local LLMs.
Events flow through a clean, composable pipeline from source to action.
Define your pipeline in YAML, point it at your data, and let Tunnlo handle the rest.
sources: - id: network-monitor adapter: native/stdin filters: - type: rate-limiter max_events_per_minute: 30 - type: windowed-aggregation window_seconds: 10 max_batch_size: 5 summary_prompt: | Analyze these network events. Flag any suspicious patterns. agents: - id: fast-triage runtime: direct-llm model: ollama/llama3.1:8b system_prompt: | Quick triage: flag or dismiss. - id: deep-analysis runtime: direct-llm model: anthropic/claude-sonnet-4-5 system_prompt: | Deep security analysis. Be concise and actionable. dashboard: enabled: true
# Install $ npm install -g tunnlo # Run with piped data $ cat events.jsonl | npx tunnlo start pipeline.yaml # Or with live network data $ npx tunnlo start pipeline.yaml # View dashboard $ open http://localhost:4400 # Check status $ npx tunnlo status ✓ Pipeline running Events: 142 received, 28 sent to LLM Uptime: 3m 22s Dashboard: http://localhost:4400
First-class support for popular LLM providers, message buses, and agent frameworks.
Tunnlo is open source and free to use. Join the community and start bridging your data to agents.