Tilion (YC F26) claims its free released Fortress v3 โ a stealth Chromium built to evade anti-bot fingerprinting โ lets web agents run at millions-of-agents scale without getting blocked; adoption as a standard evasion substrate would escalate the site-vs-agent access arms race.
state: seedheat: lowuncertainty: mediumconvergesscott: highagentic-security browser-agents anti-bot-evasion browser-infraTilionArham (Tilion CTO)
What is this?
Fortress is an open-source (BSD-3) Chromium fork from Tilion, a YC F26 startup, that spoofs browser fingerprints inside the engine โ 34 single-surface C++ patches to Blink, V8, and BoringSSL rather than an injectable JavaScript layer โ so Playwright/Puppeteer automation reads as an ordinary Chrome install and, per the founders' own captures, passes CreepJS, Sannysoft, BrowserScan, and live Cloudflare challenges. The founders released v3 free as a drop-in CDP replacement plus a beta MCP server exposing ~29 stealth-browser tools aimed specifically at AI agents (browser-use, Crawl4AI, Stagehand, LangChain integrations). The free engine is the wedge for Tilion Cloud, a hosted version with residential egress and warm session pools pitched at pointing 'thousands of agents' at the web at once; the launch reportedly drew 15k+ installs and the HN front page. The snippets show this is a forming category โ Steel Stealth Browser and Camoufox also do source-level evasion โ but all capability and scale claims here come from Tilion itself; no independent verification or anti-bot-vendor response appears in the supplied material.
Why it matters to Scott
Converges with the line Scott's own ebook chapter already argued and his Patchright stack already practices โ engine-level stealth Chromium for bot-resistant agent capture โ making Fortress an immediate evaluate-or-dismiss decision for his capture tooling and a live test of his dental-lab thesis that escaping the browser (payload capture, direct GraphQL) beats fighting fingerprint wars. It also sharpens Agent Addressability: free industrial mimicry is exactly the impersonation path the framework rules out, so adoption either delays the delegation-surface equilibrium or forces the Windi-style access-contract response โ with a defender-side echo for his own Cloudflare-fronted estate and bot-telemetry plugin, and with every capability and scale claim still Tilion's own, unverified in the supplied material.
ip:framework.agent-addressabilitydev:technology.patchrightip:source.breaking-new-ground-in-ai-agent-automation-ai-chrome-vision-and-openai-s-structured-output-ebookdev:project.browserusework:concept.wp-hosting-performance-checkradar:concept.browser-agentsradar:concept.ai-crawlersradar:windi-web-agent-access-contractradar:chromium-cve-2026-85046-sandbox-rce
queries asked of Scott's wikis
- browser agent harness Playwright Puppeteer CDP automation
- agent web access blocked anti-bot arms race site permissions
- MCP server tools agent tooling integrations
- open core free engine hosted cloud product pattern
- multi-agent fleet scale concurrency economics
- Chromium fork maintenance cost upstream rebase adoption
Measured heat
now 0 pts/hpeak 1 pts/hcomments 0/hpeers p14momentum: steady2 platformsage 232h
points/hour across evidence ยท reading as of 2026-10-12 02:59:37.977291+11:00 ยท deterministic, not a model opinion
How the heat travelled
pace: p38 vs 1188 stories at the 168h mark (now 232h old) โ ahead of agentsec-static-config-auditing (1.2x), behind agent-trace-tampering (0.8x)
Evidence (2) โ โญ canonical anchor
Interpretation history
2026-10-02T01:39:18Z
grounded: converges/high โ Converges with the line Scott's own ebook chapter already argued and his Patchright stack already practices โ engine-level stealth Chromium for bot-resistant ag
2026-10-02T01:28:12Z
case created โ First-party release by the founders themselves of purpose-built, free agent fingerprint-evasion infrastructure โ a concrete escalation in the site-vs-agent arms race distinct from the Windi governance-spec case.
Decision trace
- 10-04 04:56review_screenThe only changes are two new comments expressing moral disapproval of bot evasion; opinions without new facts, results, contradictions, or first-hand evidence do not alter the assessment of the alread
- 10-02 11:39groundConverges with the line Scott's own ebook chapter already argued and his Patchright stack already practices โ engine-level stealth Chromium for bot-resistant agent capture โ making Fortress an im
- 10-02 11:28createFirst-party release by the founders themselves of purpose-built, free agent fingerprint-evasion infrastructure โ a concrete escalation in the site-vs-agent arms race distinct from the Windi governance