TinyFish

Enterprise web-agent infrastructure: a single API providing search, fetch, browser, and multi-step agent endpoints for AI agents that operate the live web.

1. Core Product / Service

TinyFish sells "infrastructure for AI agents operating the live web". A single API key exposes four endpoints [1][2]:

Search — real-browser-rendered search, returns structured JSON
Fetch — any URL converted to markdown / JSON / HTML
Browser — Stealth Chromium session, bypasses anti-scraping, handles login state, sub-250ms cold start
Agent — multi-step automation (filling forms, auth flows, extraction)

Technical highlights: MCP-native (directly connects to Claude / Cursor), uses semantic elements in place of CSS selectors, Mind2Web benchmark 89.9% accuracy (self-reported highest public value) [2]. At its 2026-04-14 launch, positioned as "smashing four vertical tools into one platform" [4].

The "multi-IM-channel agent collaboration / role boundary / in-org agent" angle Jimmy previously focused on in raw notes — based on current public info, TinyFish's external messaging is not as an IM agent collaboration platform but as web agent infrastructure. The early IM-channel descriptions may have been an early product form or a mismemory; the current official mainline is enterprise web automation.

2. Target Users & Pain Points

Target: large enterprises, agent builders, developers needing to fetch data / interact on the live web.

Pain points:

Sites are increasingly SPA / dynamically rendered / anti-scraping, traditional scraping fails
Data is in the "hidden web" — back-office systems, login-gated portals, outdated booking systems
Stitching together search + scraper + headless browser + agent framework yourself is too heavy

Typical customer use cases:

Google — uses TinyFish agents to aggregate inventory from thousands of Japanese hotels into Google hotel search (those hotels' booking systems are too old to integrate via API directly) [3]
DoorDash — "TinyFish's platform manages web interaction complexity at scale" (Abhi Shah, Director of Data Science) [2]
Others: Cigna, Grubhub, Volkswagen, NEC, Amazon, ClassPass, The Zebra

The platform claims to have handled 40M+ agent operations with 99.99% uptime [2].

3. Competitive Landscape

TinyFish positions itself as "enterprise web agent infra", overlapping with several lanes:

OpenAI Operator / Anthropic Computer Use — general web agents; TinyFish self-reports 81% on complex web tasks vs Operator 43% [2]
Browser automation infra: Browserbase, Browserless, Steel.dev — provide hosted browsers but don't do agent orchestration
Search / scraping API: Tavily (tavily is similar in the search-API angle), Firecrawl, Bright Data — TinyFish's Search/Fetch competes directly
Enterprise agent platform: Sierra, Decagon, crescendo — these three mainly do customer service / conversational agents; TinyFish isn't on the same layer, more often it's infra sold to them or going directly to enterprise around them

Differentiation: bundling search + fetch + browser + agent into one SKU, one credit pool, one API key, lower engineering complexity than "stitching four vendors together".

4. Unique Observations

Early raw notes described TinyFish as "multi-IM-channel agent collaboration / in-org agent role boundary", which doesn't match the 2026-05 website at all. Two possibilities: (a) the company pivoted; (b) the early notes confused it with another company. Future TinyFish materials should prioritize the web agent infra line.
The combo of "Search/Fetch free, Agent/Browser usage-priced" is a classic move of making the commodity portion free as a funnel — neither Tavily nor Firecrawl dared to price this way; worth watching for the squeeze it puts on other search-API companies.
ICONIQ leading the Series A is a strong signal — ICONIQ tends to invest in later stages, so a $47M Series A in the web-agent infra lane in 2025 is a large check.
Relationship to hermes-agent / review-agent: as a counterpoint to a self-built Personal AI stack, TinyFish is the turnkey choice "if you don't want to write puppeteer / playwright / Tavily stitching yourself", but vendor lock-in and credit-based billing are unfriendly to hobbyists / small shops.

5. Financials / Funding

Total funding: $47M, 1 round [1]
Series A: 2025-08-20, $47M, ICONIQ Capital lead [1][3]
Follow-on: U.S. Venture Partners, MongoDB Ventures, Sandberg Bernthal Venture Partners, J12, etc. [1]
Founded: 2023, HQ Cupertino, CA [1]
Valuation undisclosed

Additionally: in 2026, launched "TinyFish Accelerator" with $2M from Mango Capital, funding founders building on the agentic web — a typical infra-company "build your own downstream" play.

6. People & Relationships

Founders: Shuhao Zhang, Keith Zhai, Sudheesh Nair (CEO) [1]
Lead: ICONIQ Capital
Enterprise customers: Google, DoorDash, Cigna, Grubhub, Volkswagen, NEC, Amazon, ClassPass, The Zebra
Ecosystem counterparts: crescendo (enterprise customer-service agent, different layer), hermes-agent (Jimmy's self-built agent framework), review-agent (Jimmy's self-built workflow)

Sources

[1] BusinessWire — TinyFish launches with $47M (2025-08-20) — https://www.businesswire.com/news/home/20250820555825/en/ (accessed 2026-05-09)
[2] tinyfish.ai website (2026-05-09)
[3] ICONIQ partnership post (2026-05-09)
[4] MarkTechPost — TinyFish 4-in-1 platform release (2026-04-14)
local: 2026-04-27-summary.md (referenced by original wiki, not found in raw/, uncertain)