Back to vendors
N

Notte

Visit site
Browser / computer-use agentindependentVerified 2026-07-01

Open core full stack framework and browser infrastructure for building, deploying, and scaling reliable web agents, using a perception layer that turns webpages into natural language action maps.

Notte, a Y Combinator company from Notte Labs, is an open core, full stack framework and browser infrastructure for building, deploying, and scaling reliable web agents through a single application programming interface. Its signature idea is a perception layer that transforms webpages into structured, navigable maps described in natural language, an action interface in text. This lets language models read, understand, and act on sites without fragile CSS or XPath selectors, and reduces their cognitive load enough that smaller, faster, cheaper models can drive accurate automation. The framework pairs AI agents with traditional scripting so developers script deterministic parts and use AI only where needed, cutting cost while improving reliability.

As infrastructure, Notte ships managed cloud browser sessions on a global edge network for low latency, plus serverless browser functions that colocate compute next to the browser, described as AWS Lambda for browser work. It handles the operational hard parts out of the box: proxies, CAPTCHA solving, anti detection, browser profiles, and an encrypted secrets vault where agents fetch credentials at runtime rather than hardcoding them. It is compatible with Playwright, Puppeteer, Selenium, browser-use, Stagehand, and anything that connects over the Chrome DevTools Protocol, in Python or TypeScript.

Notte is model agnostic, working with OpenAI, Anthropic, Gemini, and others, and developers can bring their own keys with token pricing passed through at no markup or select the reasoning model per agent. Extensibility is a strength: alongside the open source framework and software development kits, Notte publishes a Model Context Protocol server that connects language models directly to its cloud browser. For observability it offers live view screen sharing, session replays, live debugging in a browser based studio, and production monitoring, and it publishes open browser agent benchmarks like open operator evals.

The platform is SOC 2 Type II certified and can run as a managed cloud service or be self hosted from its source available codebase, and it supports bringing an external browser provider over the Chrome DevTools Protocol. Its main current gaps are that a first class persistent agent memory and a shipped human in the loop workflow remain on the roadmap, and it is a developer framework rather than a marketplace of prebuilt agents. For engineering teams building production web automation, it is a fast, cost efficient, and well instrumented foundation.

Vendor details

Canonical URL

https://notte.cc

Category

Browser / computer-use agent

Company status

independent

Use cases & customers

In practice

A developer builds a flight search agent. Using Notte's perception layer, the agent reads Google Flights as a natural language action map, fills the form, and runs the search reliably with a small model, deployed as a serverless function.

A data team needs structured extraction at scale. They call Notte's scraping endpoint with a schema, turn target pages into machine readable JSON, and run it on managed cloud browsers with proxies and CAPTCHA solving handled automatically.

An engineering team wants an agent driven from Claude. They connect through Notte's Model Context Protocol server, letting the assistant execute web tasks on Notte's cloud browsers while credentials stay in the encrypted secrets vault, never hardcoded.

Sources & related URLs

Research notes

Open-core full-stack WEB AGENT FRAMEWORK + browser infrastructure. Notte Labs, Inc. YC S25. Founders Andrea Pinto, Lucas Giordano. SSPL-1.0 license (source-available/self-hostable) + cloud-hosted. SIGNATURE: PERCEPTION LAYER — transforms webpages → structured navigable maps in natural language (action-API-in-text), eliminates fragile CSS/XPath selectors, reduces LLM cognitive-load → use smaller/faster/cheaper models. Observe/Step/Scrape primitives (Playwright-compatible + NL). Combines AI-agents + traditional-scripting (script deterministic parts, AI only when needed — cuts cost 50%+). Single API, Python + TypeScript SDK (notte_sdk). INFRA: managed cloud browser sessions (headless+headful) on global edge network (zero-latency), serverless browser functions ('AWS Lambda for browser work', colocate compute), CDP/Playwright/Puppeteer/Selenium/browser-use/Stagehand compatible. Stealth: CAPTCHA-solving, residential-proxies, anti-detection. Secrets Vault (encrypted, runtime-fetch, never-hardcoded) + browser-profiles + personas. 6 FULLS: Sec=F (SOC 2 Type II CERTIFIED [verified on FAQ] + enterprise-grade encrypted secrets-vault + web-security), Obs=F (live-view screen-sharing + session-replays + live-debugging Studio/browser-IDE + production-observability + console-monitoring + debug-endpoint), Dep=F (OPEN-CORE self-hostable [SSPL] + managed-cloud + edge + BYO-browser-over-CDP), Model=F (MODEL-AGNOSTIC: OpenAI/Anthropic/Gemini/others + BYOK token-no-markup + reasoning_model per-agent), Ext=F (single-API + Python/TS SDK + open-source framework + MCP SERVER [Notte Browser MCP, connects LLMs/Claude/Cursor to cloud-browser] + CDP-compat), Comp=F (perception-layer web agents navigate/act autonomously, handle dynamic-sites/CAPTCHA/anti-bot, 'operate web like real users', SOTA browser-agent benchmarks). 5 P's: Int=P (any-site + Playwright/Puppeteer/Selenium/ecosystem interop; not native-app-connectors), Orch=P (goal→autonomous multi-step nav+extraction + AI+scripting; not multi-agent), Know=P (perception-layer structured-maps + SOTA structured-schema scraping/JSON-extraction), Trig=P (serverless functions invoke/schedule anywhere via API), Eval=P (live-debugging + session-replays + publishes open benchmarks open-operator-evals/browserbench/browserarena; framework-level not per-user-agent-harness). 3 N's: HITL=N (deterministic-scripting control + HITL-integration is ROADMAP; not shipped), Mem=N (cookies/profiles/session-state + 'improved memory' is ROADMAP; no shipped agent-memory), Pack=N (framework + examples + cookbook-coming; no prebuilt-agent marketplace). Benchmarks: 'outperforms browser-agents in all benchmarks'; independent (skywork) confirms leads Browser-Use on speed/reliability. Roadmap: captcha/residential-proxies/web-security/VPN-browsing/auth+payments/improved-memory/HITL/notifications/cookies. Strongest net-new browser-lane infra play. Pricing: CREDIT-BASED (credits on sessions/functions/proxies/LLM-tokens), large FREE tier (100 browser hours), BYO-keys token-no-markup, open-source free self-host → self_serve. Domain notte.cc. Score 8.5 (6F/5P/3N).

Capability coverage

8.5 / 14 capabilities · 61%

Integrations & Tool CallingNotte agents act across any website and the framework integrates with Playwright, Puppeteer, Selenium, and the broader browser tooling ecosystem through a single application programming interface, real integration short of a documented library of native business application connectors, so partial. Partial
Workflow OrchestrationNotte agents take a goal and autonomously handle multi step web navigation and extraction, combining AI steps with deterministic scripting, real orchestration short of documented multi agent coordination, so partial. Partial
Knowledge Grounding & RAGNotte's perception layer turns webpages into structured natural language maps and its scraping endpoints extract structured data, strong grounding in web content short of a documented citation grounded retrieval system, so partial. Partial
Human Oversight & GuardrailsNotte lets developers script deterministic parts for control and lists human in the loop integration as an in progress feature, but a shipped runtime human oversight or guardrail enforcement engine could not be verified. Unable to verify
Security, Identity & GovernanceNotte is SOC 2 Type II certified and provides an enterprise grade encrypted secrets vault where agents fetch credentials at runtime rather than hardcoding them, plus web security features, so full. Full
Observability & AuditabilityNotte provides live view screen sharing, session replays, live debugging in a browser based studio, and production observability through its console, so full. Full
Memory & State PersistenceNotte manages cookies, browser profiles, and session state and lists improved memory as an in progress feature, but a shipped persistent cross session agent memory could not be verified. Unable to verify
Deployment & Data ResidencyNotte is open core and self hostable from its source available codebase, runs as managed cloud sessions on a global edge network, and supports bringing your own browser provider over the Chrome DevTools Protocol, so full. Full
Prebuilt Agents, Templates & PacksNotte is a framework for building custom agents with example recipes and a forthcoming cookbook, but a library of prebuilt or cloneable agents could not be verified. Unable to verify
Triggers & Channel CoverageNotte deploys agents and automations as serverless functions that can be invoked or scheduled anywhere through its application programming interface, real triggering short of broad customer facing channel coverage, so partial. Partial
Model Flexibility & RoutingNotte is model agnostic, working with OpenAI, Anthropic, Gemini, and others, and lets developers bring their own keys and select the reasoning model per agent, so full. Full
APIs, SDKs & MCP ExtensibilityNotte offers a single application programming interface, Python and TypeScript software development kits, an open source framework, and a Model Context Protocol server that connects language models directly to its browser infrastructure, so full. Full
Testing, Debugging & OptimizationNotte provides live debugging and session replays and publishes open browser agent benchmarks like open operator evals, real testing and evaluation tooling short of a documented in product evaluation harness for user agents, so partial. Partial
Browser & Computer UseNotte's web agents navigate and act on websites autonomously through a perception layer that turns pages into action maps, handling dynamic sites, CAPTCHAs, and anti bot defenses, so full. Full

Recent platform changes

No recent material changes tracked yet.

Pricing

Free tier (100 browser hours); usage based credit plans

Public — exactHigh variable cost
Verified 2026-07-01

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.

Agentic AI Index

A directory and comparison resource for AI agent platforms, autonomous workflow tools, and enterprise agentic automation products.

© 2026 Agentic AI Index

3801 N Capital of Texas Hwy, Ste E240 · Austin, TX 78746

Researched from public vendor sources. See Methodology.