Lightpanda
Also known as: Lightpanda Browser, lightpanda-io, lightpanda.io
A headless browser engine written from scratch in Zig for machines, CDP compatible with Playwright and Puppeteer, with a native MCP server, a built in agent on the LLM of your choice, deterministic PandaScripts, and a self hosted binary or managed cloud in EU and US regions.
Lightpanda is a headless browser engine written from scratch in Zig with V8 for web automation, scraping and AI agents. It is not a fork of Chromium, and it runs JavaScript and the DOM without a rendering pipeline. Lightpanda reports it about ten times faster and lighter than headless Chrome. It speaks the Chrome DevTools Protocol and WebDriver BiDi, so Playwright, Puppeteer, chromedp and Selenium connect to it, and it ships a native MCP server with a published tool set, an HTTP fetch API returning HTML or Markdown, a Python SDK and a CLI.
It also includes Lightpanda Agent, which drives the browser from plain English or slash commands on the LLM provider the user chooses, including a local model, and can save a session as a PandaScript, a deterministic script Lightpanda replays with no LLM. The engine is open source and runs on the customer's own machines, where no traffic reaches Lightpanda, or on Lightpanda Cloud with EU and US endpoints, a session console, managed proxies and usage based plans. Robots.txt compliance is opt in.
Vendor details
Canonical URL
https://lightpanda.io
Category
Agent infrastructure
Funding status
Not stated on the pages read; the engine is open source (GitHub lightpanda-io) with a commercial managed cloud
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
CDP and WebDriver BiDi connect Playwright, Puppeteer, chromedp and Selenium, and guides cover Stagehand and Hermes Agent. Lightpanda also offers a native MCP server (stdio or HTTP, also on Cloud), an HTTP fetch API, a Python SDK and a CLI.
In practice
Your team's AI agent needs to browse the web through a standard tool interface. Lightpanda's native MCP server gives it navigation, extraction, interaction and session tools, locally or on Lightpanda Cloud.
Your team has a browsing task the agent handles well and needs it to run the same way every time. Lightpanda's built in agent saves the session as a PandaScript, which Lightpanda replays without an LLM.
Your team's scraping cannot send pages or data to an outside service. Lightpanda runs as a self hosted binary on your own machines, and your existing Playwright or Puppeteer code connects to it over CDP.
Sources & related URLs
Agentic Index coverage score
6.0 / 14 capabilities · 43%
| Integrations & Tool Calling | Partial |
|---|---|
|
Guides connect Lightpanda to Stagehand, Hermes Agent, Playwright, Puppeteer, chromedp and Selenium, all frameworks that drive the browser. The agent's own tools are browser tools plus web search. Lightpanda names no authenticated action route, credential handling or connector for acting in outside systems. Sourcelightpanda.io/docs/guides/use-stagehandread 2026-09-27 |
|
| Workflow Orchestration | Partial |
|
The built in agent carries out multi step browsing tasks and can save a session as a PandaScript, a deterministic JavaScript automation with loops and functions that Lightpanda replays without an LLM. Lightpanda describes no multi agent execution and no workflow model beyond a single script. Sourcelightpanda.io/docs/usage/pandascriptread 2026-09-27 |
|
| Knowledge Grounding & RAG | Not documented |
|
The agent extracts pages and searches the web at run time. Lightpanda names no knowledge grounding or retrieval layer. Sourcelightpanda.io/docs/usage/agentread 2026-09-27 |
|
| Human Oversight & Guardrails | Partial |
|
The customer can set --obey_robots so every request the browser makes, including for an agent driving it, is checked against the site's robots.txt and blocked where disallowed. This lets the customer limit what the agent may do. Lightpanda names no approval step before an agent action commits. Sourcelightpanda.io/blog/posts/robotstxt-supportread 2026-09-27 |
|
| Security, Identity & Governance | Not documented |
|
Access to Lightpanda Cloud runs on a single API token per account, with no roles, SSO or team access model. No attestation or trust page is published. The Enterprise plan offers support for advanced security needs without naming them. Sourcelightpanda.io/pricingread 2026-09-27 |
|
| Observability & Auditability | Partial |
|
The Lightpanda Cloud console lists recent sessions with their logs, and the local binary writes request, navigation and resource logs at a chosen level to stderr. Lightpanda names no replay, trace export or audit trail of agent actions. Sourcelightpanda.io/docs/run-on-lightpanda-cloud/getting-startedread 2026-09-27 |
|
| Memory & State Persistence | Not documented |
|
PandaScript variables last for one script run, and an MCP session keeps its page loaded between calls. The agent remembers the last chosen provider for each directory as a setting. Lightpanda names no memory that the agent reads across runs. Sourcelightpanda.io/docs/usage/pandascriptread 2026-09-27 |
|
| Deployment & Data Residency | Full |
|
The same engine runs as a self hosted binary on the customer's own machines or infrastructure, where pages and data never reach Lightpanda, or on Lightpanda Cloud with euwest and uswest endpoints the customer picks. Enterprise offers on premises or private cloud deployment. Sourcelightpanda.io/docs/core-concepts/local-vs-cloudread 2026-09-27 |
|
| Prebuilt Agents, Templates & Packs | Not documented |
|
PandaScripts are written or saved by the customer, and Lightpanda names no catalog of prebuilt agents, templates or scripts. Sourcelightpanda.io/docs/usage/pandascriptread 2026-09-27 |
|
| Triggers & Channel Coverage | Not documented |
|
Work starts when a caller connects over CDP, MCP or the HTTP API, or when a user runs the agent or a PandaScript from the command line. Lightpanda names no schedule, event or inbound trigger of its own. Sourcelightpanda.io/docs/reference/cliread 2026-09-27 |
|
| Model Flexibility & Routing | Full |
|
The built in agent runs on the provider the customer chooses, set by API key, --provider flag or /provider command. The choices are Anthropic, OpenAI, Google, Hugging Face, Mistral, OpenRouter, OrcaRouter and the Vercel AI Gateway, or a local model through Ollama or llama.cpp. Sourcelightpanda.io/docs/usage/agentread 2026-09-27 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Developers can call an HTTP API (POST /api/fetch with a bearer token), a Python SDK, a CLI and a CDP server. A native MCP server over stdio or HTTP offers navigation, extraction, interaction and session tools that drive Lightpanda's own browser sessions, and it is also exposed on Lightpanda Cloud with a token. Sourcelightpanda.io/docs/reference/mcp-toolsread 2026-09-27 |
|
| Testing, Debugging & Optimization | Not documented |
|
Published benchmarks measure the engine's speed and memory use, and Lightpanda names no evaluation harness, test cases or quality gate for agents. Sourcelightpanda.io/docs/core-concepts/benchmarksread 2026-09-27 |
|
| Browser & Computer Use | Full |
|
Lightpanda is its own headless browser engine, written in Zig with V8, that runs JavaScript and the DOM. Agents drive it over CDP or WebDriver BiDi (Playwright, Puppeteer, chromedp), through its MCP server, or with the built in agent's goto, click, fill and extract tools, locally or on Lightpanda Cloud. It skips visual rendering, so there are no screenshots. Sourcelightpanda.io/docs/usage/agentread 2026-09-27 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Lightpanda's headless browser for AI agents reached version 1.0 and dropped its beta label, after passing about 80% of the Web Platform Tests that Chrome passes. The release enforces CORS by default on every request a page makes, so a hostile page can no longer read internal services from inside the network.
Bears on: Security / enterprise
View sourcePricing
Open source engine free to self host; Cloud Explorer free (10 browser hours a month); Builder $19 a month (300 hours, then $0.08 an hour); Enterprise custom
browser hours metered by the second
What is public
lightpanda.io/pricing publishes the Explorer and Builder plans with hour pools, concurrency and the overage rate; Enterprise is quoted.
Variable cost rationale
The product's core pitch is collapsing per instance resource cost (roughly 9x lower memory than Chrome), so browser fleet spend scales gently with volume whether self hosted or cloud.
Overage / add-ons
Builder keeps running past its 300 included hours and bills $0.08 per browser hour; the free plan stops until its pool resets.
Sales call required
No, self serve available
Free / trial
Open source self hosting free; Cloud Explorer plan free with 10 browser hours a month
Lowest paid plan
Builder, $19 a month for 300 browser hours and 30 concurrent sessions, then $0.08 an hour
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
- Apify — Cloud platform and marketplace of more than 73,000 ready-to-run…
Alternatives to Lightpanda
The closest documented capability profiles to Lightpanda among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Hyperbrowser8.0 / 14Adds documented Security, Identity & Governance and Memory & State Persistence, among others
- GPT Researcher7.0 / 14Adds documented Knowledge Grounding & RAG and Prebuilt Agents, Templates & Packs, among others
- Jina AI5.0 / 14Adds documented Knowledge Grounding & RAG and Security, Identity & Governance
- Rye5.0 / 14Fuller documented coverage on Integrations & Tool Calling and Human Oversight & Guardrails
- Vocode6.0 / 14Adds documented Memory & State Persistence and Triggers & Channel Coverage
- Circuit & Chisel4.5 / 14Adds documented Memory & State Persistence
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded