Back to vendors

Tinyfish

Visit site

Browser / computer-use agentindependentVerified 2026-07-01

Enterprise infrastructure for AI web agents that navigate, authenticate, extract, and transact across hundreds of live sites in parallel through a single Model Context Protocol native API.

TinyFish is enterprise infrastructure for AI web agents, founded in 2024 in Palo Alto and led by chief executive Sudheesh Nair, and it launched with forty seven million dollars in funding led by ICONIQ Capital. The premise is that much of the web's most valuable data sits behind logins, forms, and paywalls where search engines cannot reach and traditional automation fails. TinyFish deploys swarms of specialized agents that sign in, navigate, extract, and act like humans at scale, and already runs hundreds of thousands of enterprise web agents every month for customers including Google, DoorDash, Amazon, and ClassPass across hospitality, insurance, retail, and logistics.

The platform brings four capabilities under one application programming interface key and one credit pool: Search returns structured results from live pages rendered by real browsers rather than cached indexes, Fetch converts any URL into clean markdown or JSON, Browser provides stealth sessions that defeat anti bot protection, and Agent runs autonomous multi step web workflows. It is serverless, with no browsers to manage or proxies to configure, and can run up to a thousand simultaneous operations across hundreds of sites, completing in minutes what would take days by hand.

TinyFish is Model Context Protocol native, so it works with Claude, Cursor, and any compatible client, and it also offers a direct application programming interface plus integrations with tools like n8n and Vercel. Its open source AgentQL project adds an AI powered query language with Python and JavaScript software development kits, a REST endpoint, and a browser based debugger, using natural language selectors that self heal as sites change. Agents preserve state and authentication across sessions and return structured outputs to downstream systems, webhooks, and audit trails.

The platform is built for enterprise governance, emphasizing consistency, inspectability, and control through deterministic web execution and structured, auditable artifacts. In its own framing it reports an eighty one percent success rate on complex web tasks against a much lower figure for a leading consumer agent. Its main gaps for the index are that named security certifications, a self hosted deployment of the full platform, a persistent learning memory, and user facing model selection were not clearly documented, since it deliberately bundles model inference into one usage based price.

Vendor details

Canonical URL

https://tinyfish.ai

Company status

independent

Use cases & customers

In practice

An energy company must pull invoices from hundreds of vendor portals. TinyFish agents log into 341 portals across 1,643 monthly logins, navigate multi step workflows, and return structured invoices and status updates to downstream systems automatically.

A small hotel in Japan with eight rooms and no API needs its availability on Google. A TinyFish agent signs into the booking system, reads dates and pricing, and updates the Google Hotels listing without the hotel changing anything.

An insurance marketplace collects quotes from many carriers. TinyFish handles each provider's different authentication and form structure, retrieves rates, and returns structured, comparable quotes, running many portals in parallel in a fraction of the manual time.

Sources & related URLs

Research notes

Enterprise INFRASTRUCTURE for AI web agents. Palo Alto. Founded 2024. $47M Series A (ICONIQ Capital lead, Mango Capital). CEO Sudheesh Nair (ex-Nutanix president, ex-ThoughtSpot CEO — experienced enterprise exec). Marquee customers: Google, Amazon, DoorDash, ClassPass, Grubhub, Fortune 500 (hospitality/insurance/retail/logistics). 'Schools of fish' — thousands of specialized agents SWARM. Runs 100,000s+ enterprise web agents/month. FOUR products, ONE API key, ONE credit pool: Search (live web → structured JSON from real browsers, not cached), Fetch (URL → clean markdown/JSON/HTML token-efficient), Browser (stealth cloud sessions), Agent (autonomous multi-step web workflows: navigate/fill-forms/authenticate/transact/structured-results). SERVERLESS (no browsers/proxies to manage, single API), up to 1,000 PARALLEL operations across hundreds of sites, sub-250ms cold-starts, persistent-state. Stealth (login-walls/anti-bot/dynamic-rendering, residential-proxies, CAPTCHA). SEMANTIC element-finding (self-healing, not CSS-selectors). Live-web-data (behind logins/forms/paywalls, minutes-fresh). AgentQL (tinyfish-io/agentql, OPEN-SOURCE: AI-query-language + Python/JS SDKs + browser-debugger + REST-API + Langchain/Zapier/MCP + playground). Voyage-AI embeddings + MongoDB (product-matching/retrieval). 4 FULLS: Int=F (MCP-native Claude/Cursor + Cursor/n8n/Dify/OpenClaw/Vercel/Langchain/Zapier/Python/Node + web-actions + webhooks/downstream), Orch=F (autonomous agent-swarm + state/auth-preserved + 1,000-parallel across hundreds-of-sites), Ext=F (single-API + MCP-native + open-source AgentQL Python/JS-SDK/REST/debugger), Comp=F (stealth vision/semantic browser nav, authenticate/fill-forms/transact like-human, anti-bot/dynamic-rendering). 8 P's: Know=P ('AI Agent Web Grounding' live-web-context-for-LLMs + Search/Fetch + Voyage embeddings/reranking), HITL=P (inspectability/control/governance/deterministic; no runtime-approval), Sec=P (corporate-security+governance + audit-trails + Google/Amazon-customers; NO NAMED CERTS surfaced — likely SOC 2 exists but unverified, flag for later pass), Obs=P (structured-artifacts + audit-trails + inspectability + AgentQL-debugger; not full-trace/replay), Mem=P (persistent-state + auth-preserved-across-sessions; not learned-memory), Dep=P (serverless-cloud + open-source-AgentQL-tooling; core-platform cloud-only, no self-host), Trig=P (API + MCP + n8n + webhooks + monitoring-schedules), Eval=P (AgentQL-debugger + playground + benchmarks [81% vs Operator 43%] + QA-testing-support). 2 N's: Pack=N (specialized-per-use-case agents, no prebuilt-agent marketplace), Model=N (BUNDLES own LLM-inference into one price, no customer model-selection — deliberate abstraction). Deterministic-execution + audit-trails + governance = enterprise-grade. Note: like Emergence, arguably broader 'web-infra' than pure browser-agent, but Agent product IS browser-agent — category fits. Pricing: TRANSPARENT usage-based, ALL costs bundled (LLM + browsers + proxies + anti-bot); free 500-steps (no card); Starter $15/mo; usage-based + enterprise → self_serve. Domain tinyfish.ai. Score 8.0 (4F/8P/2N).

Capability coverage

8.0 / 14 capabilities · 57%

Integrations & Tool CallingTinyFish is Model Context Protocol native and integrates with Cursor, n8n, Langchain, Zapier, Python, and Node.js while its agents authenticate into and act across hundreds of live sites, returning structured outputs to downstream systems and webhooks, so full.	Full
Workflow OrchestrationTinyFish deploys swarms of autonomous agents that execute multi step web workflows with state and authentication preserved, running up to a thousand operations in parallel across hundreds of sites, so full.	Full
Knowledge Grounding & RAGTinyFish grounds language model agents in live web data through its Search and Fetch products and uses embedding and reranking models for semantic retrieval and product matching, strong grounding short of a documented citation grounded answer system, so partial.	Partial
Human Oversight & GuardrailsTinyFish emphasizes inspectability and control with deterministic web execution and governance for enterprise workflows, real control short of a documented runtime human approval or guardrail enforcement engine, so partial.	Partial
Security, Identity & GovernanceTinyFish provides corporate level security and governance with audit trails and is deployed by enterprises like Google and Amazon, a strong enterprise posture short of documented named certifications like SOC 2 with role based access, so partial.	Partial
Observability & AuditabilityTinyFish produces structured artifacts and audit trails and emphasizes inspectability and observable outcomes, plus a browser debugger in its AgentQL tooling, real observability short of a documented full per action trace and replay system, so partial.	Partial
Memory & State PersistenceTinyFish preserves persistent state and authentication across sessions for its agents, real state persistence short of a documented general cross session agent memory that learns over time, so partial.	Partial
Deployment & Data ResidencyTinyFish runs as a serverless cloud platform with no infrastructure to manage, and its AgentQL query language and software development kits are open source, real deployment flexibility short of a documented self hosted or on premises agent platform, so partial.	Partial
Prebuilt Agents, Templates & PacksTinyFish deploys specialized agents per enterprise use case and offers example tooling, but a browsable library of prebuilt or cloneable agents could not be verified.	Unable to verify
Triggers & Channel CoverageTinyFish agents are triggered through its application programming interface, Model Context Protocol, and integrations like n8n, returning outputs to webhooks and downstream systems, real triggering short of broad customer facing channel coverage, so partial.	Partial
Model Flexibility & RoutingTinyFish bundles its own large language model inference into a single price and uses embedding models like Voyage AI, but user facing multi provider model selection or routing could not be verified.	Unable to verify
APIs, SDKs & MCP ExtensibilityTinyFish exposes a single application programming interface, is Model Context Protocol native, and open sources its AgentQL query language with Python and JavaScript software development kits, a REST API, and a browser debugger, so full.	Full
Testing, Debugging & OptimizationTinyFish provides a browser debugger and playground to test and optimize queries and publishes benchmark comparisons, and supports building QA testing agents, real testing tooling short of a documented in product evaluation harness for agent workflows, so partial.	Partial
Browser & Computer UseTinyFish agents navigate, authenticate into, fill forms on, and transact across live websites like humans using stealth browser sessions and semantic element finding, handling anti bot defenses and dynamic rendering, so full.	Full

Recent platform changes

No recent material changes tracked yet.

View all changes for Tinyfish →

Pricing

Free Search/Fetch; PAYG $0.015/step; Starter $15/mo, Pro $150/mo; Enterprise custom

Public — exactHigh variable cost

Verified 2026-07-01

Data confidence: high