Plano
Plano is an AI-native proxy and data plane for agentic apps — with built-in orchestration, safety, observability, and smart LLM routing so you stay focused on your agents core logic.
Last verified:
What is Plano?
Plano is an AI-native proxy and data plane for agentic applications, designed as delivery infrastructure for teams building AI agents. It handles critical plumbing work including agent routing and orchestration, rich agentic traces and observability, guardrail hooks for safety and moderation, and smart model routing APIs for LLMs. Plano is framework-friendly and protocol-native, allowing developers to use any programming language and AI framework while shipping agents to production faster.
Key features include multi-agent orchestration without framework lock-in, context engineering with reusable filters, production signals for reinforcement learning and continuous improvement, centralized security with built-in guardrails, and on-premises deployment for full data control in regulated environments. Plano is built on Envoy Proxy and includes a programmable architecture with a simple YAML configuration file. It provides Agentic Signals for zero-code capture of behavior signals, traces, and metrics across every agent.
Plano is designed for developers who want to focus on core agent product logic rather than infrastructure plumbing, product teams that need to accelerate feedback loops for reinforcement learning, and engineering teams that need to standardize policies and access controls across agents and LLMs. It is built by core contributors to Envoy Proxy and is 100% open-source under Apache 2.0 license.
Plano pricing
Pricing model: Freemium
Plano is 100% open-source under Apache 2.0 license and completely free to use. There are no paid plans or subscription tiers. Plano and the Plano family of LLMs (like Plano-Orchestrator) are hosted free of charge in the US-central region for developer experience. The cost savings come from smart model routing that can cut LLM costs by up to 50% by routing simple queries to cheaper models and complex tasks to powerful ones. Users only pay for their own LLM provider API usage (OpenAI, Anthropic, DigitalOcean, etc.).
Plano pros
- 100% open-source under Apache 2.0 license
- Framework-friendly - works with any AI framework
- Language-agnostic - use any programming language
- Built on proven Envoy Proxy infrastructure
- No GPU required - runs alongside app servers
- Simple YAML configuration file
- Multi-agent orchestration without lock-in
- Built-in guardrails for jailbreak protection
- Centralized observability with Agentic Signals
- Smart LLM routing based on prompt complexity
- On-premises deployment for data control
- Zero-code trace and metrics capture
- Reduces LLM costs by up to 50%
- Production-grade foundation for agentic apps
- Pre-compiled binaries download automatically
- Supports both native and Docker deployment
- OpenAI-compatible chat completions API
- Context engineering hooks for smarter agents
- Low-latency routing and agent handoff
Plano cons
- Relatively new project with evolving API
- Requires Python 3.10+ for CLI tool
- Learning curve for orchestration configuration
- Documentation still maturing (v0.4.x)
- Self-hosted - no managed cloud service
- Requires setup of API keys for LLM providers
- Docker mode requires Docker 24+ and Docker Compose 2.29+
- Tracing data lives in memory only by default
- Best suited for teams already using agents
Frequently asked questions about Plano
What is Plano?
Plano is delivery infrastructure for agentic apps - an AI-native proxy server and data plane designed to help you build agents faster and deliver them reliably to production. It handles the hidden AI middleware including agent routing and orchestration, rich agentic signals and traces for continuous improvement, guardrail filters for safety and moderation, and smart LLM routing APIs.
Is Plano open source?
Yes, Plano is 100% open-source under the Apache 2.0 license. The source code is available on GitHub at github.com/katanemo/plano. This means you can use it free of charge, self-host it, modify it, and contribute to its development.
What programming languages does Plano support?
Plano is language-agnostic and works with any programming language. You can implement your agents in Python, JavaScript/TypeScript, or any other language that exposes HTTP services. Plano itself is built on Envoy (C++) with a Python-based CLI tool called planoai.
How does Plano orchestrate multiple agents?
Plano uses Plano-Orchestrator, powered by small LLMs (1-4B parameters) trained for constrained routing and orchestration, to decide which agent or LLM should handle each request and in what sequence. You configure agents in a YAML file with their IDs, URLs, and descriptions, and the orchestrator analyzes prompts to route them to the right agent based on intent.
What guardrails does Plano provide?
Plano provides built-in guardrails including jailbreak detection and blocking, content policies and moderation filters, and context workflows like rewriting, retrieval, and redaction. These are applied via filter chains in the dataplane, allowing you to enforce safety and policies once centrally rather than in every agent's code.
Can I deploy Plano on-premises?
Yes, Plano supports on-premises deployment for full data control in regulated environments. It runs alongside your app servers in the cloud, on-premises, or in local dev, and doesn't require a GPU. Your GPUs stay where your models are hosted while Plano handles the routing and orchestration.
How do I get started with Plano?
To get started, install the planoai CLI using uv tool install planoai or pip install planoai. Create a plano_config.yaml file defining your LLM providers, agents, and listeners. Then run planoai up plano_config.yaml to start Plano. On first run, it automatically downloads Envoy, WASM plugins, and brightstaff to ~/.plano/.
What LLM providers does Plano support?
Plano supports any LLM provider through its model proxy. You configure providers in your config file including OpenAI, Anthropic, DigitalOcean, and others. It provides a unified OpenAI-compatible chat completions API interface so you can standardize access to multiple LLM providers through a single gateway.
How does Plano help with observability?
Plano provides Agentic Signals for zero-code capture of behavior signals, traces, and metrics across every agent. It ships two CLI tools: planoai obs for live console visibility into LLM traffic with cost tracking, and planoai trace for single-request traces showing which model was picked, hop timing, and upstream responses.
What is the difference between Plano's model proxy and agent orchestration?
The model proxy standardizes access to multiple LLM providers with a unified API interface, handling things like retries, timeouts, and connection pooling. Agent orchestration decides which agent or LLM should handle each prompt and in what sequence, using Plano-Orchestrator to route requests based on intent across your custom HTTP service agents.