Ophiris (@OphirisLoop) header
Ophiris (@OphirisLoop) profile picture

The gateway for raw intelligence: routing unrestricted LLMs across offensive security and red teaming. $OPHIRIS : 0x78c1c2d4cd6207b6ff4e84362d314d1cd0830041

Joined November 2026
Pages whisper. Now we listen. Two additions to Ophiris Recon. JS Surface Miner reads every page crawl_site and http_security already fetched. It pulls endpoints from three call shapes, keys-by-name as redacted previews (same rules as Secret Triage), and the file names a source map gives away. All of it feeds probeList(): a dir_probe wordlist with endpoints first, numeric IDs folded to [id], and a hard cap of 200 paths. Zero extra requests. On our own homepage it found /api/treasury/agent among 6 endpoints and built a 17-path list. Executor watch mode: run node executor.mjs --watch and your machine picks up jobs while the recon run is still going. It polls every 2 seconds when idle and backs off to 60 seconds on errors. In production today, 11 executor jobs came back accepted on the first try, each 2 to 12 seconds after it was queued, before the run had finished. ophiris.xyz/app
11
2
15
491
The Executor Protocol is live in Ophiris Recon. Four new checks run on your own machine: tcp_probe, tls_grab, dir_probe and sweep_ports. The CLI uses only Node's net and tls modules, so there's nothing to install. Every request is signed by your device: a key shown once, a fresh nonce, a timestamp that has to land within 5 minutes, and an HMAC-SHA256 over the request and its body. Replay a nonce and you get a 401. Two executors can never take the same job. We ran it in production. The CLI picked up a tls_grab job for ophiris.xyz, sent back the real certificate (Let's Encrypt, valid to Feb 11 2027), and ingest accepted it into the hash-chained evidence log. Eight jobs done, then we revoked the test device. Ophiris plans and records. Your machine is the only thing that touches the target. ophiris.xyz/app
2
6
289
Congrats to @AnthropicAI. $ANT was tokenized on @RobinhoodCrypto chain the same day, so we asked Ophiris to read it. One prompt. The model found the listing on its own, then pulled the quote, the on-chain pool and today's chart: • The token tracks the share: Uniswap v4 pool at $97.03 against $96.78, a 0.26% gap • Pool depth: $364.5K liquidity, $1.38M traded in 24h • First session: $107.47 to $96.78, down 9.95% at the time of the run Real session on GLM 5.3 Uncensored, 26 seconds from prompt to answer, data as of 18:32 UTC. The same tools cover all 204 tokenized assets on Robinhood Chain. Not financial advice. ophiris.xyz/app
Anthropic $ANT is live for launch pairing following its IPO. Launch and earn fees in $ANT on Pons.
1
4
263
52,873,960 OPHIRIS burned so far, 5.29% of the 1B supply. That's up 81% from our last report (17 Nov). New this week: • Secret triage in Recon finds leaked API keys in exposed files, shows them redacted, and retests to confirm they've been removed • Retest and monitoring: prove a fix, catch changes, see how attacks chain • The Recon agent, with scoped engagements and free infra checks • Light and dark mode across the site and the app These sit alongside what was already live: 68 models, coding agents, Pay with $OPHIRIS, Ouro, daily holder credits and the v4 pool. ophiris.xyz/app
1
1
5
258
An open /.env tells you a door is unlocked. It doesn't tell you what's inside. We just shipped Secret Triage inside Ophiris Recon: — 15 vendor credential signatures scanned in-memory on the bodies exposure_check already fetches (0 extra requests) — Redacted previews only (max 24 chars; test keys, doc samples & placeholders dropped) — Paired secrets_removed retest to verify cleaned-in-place files — No domain-ownership gate blocking single-GET active scope Tested live: 15/15 signatures caught in loopback, 0 false positives across 53 live hosts. Run it in your active engagements: ophiris.xyz/app
2
5
254
Find it. Fix it. Prove it. @HeyResearch had DMARC at p=none, so nothing stopped forged mail sent in their name. They moved it to reject. On 19 Nov we retested their engagement: 9 findings checked, 3 fixed, 6 still open. Forged mail in their name now bounces. New in Ophiris Recon: • Retest any finding. Free checks with no model call, and the evidence is hash-chained. • Scheduled monitoring that runs from your own device, so the engagement never leaves it. • Attack paths that show how small findings chain together, and which single fix breaks the most chains. The numbers come from the real engagement. ophiris.xyz
1
2
6
300
We pointed Ophiris Recon at heyresearch.xyz this week. Passive only, a human approved the plan, 20 tool calls, $0.15. Here’s what it found, @HeyResearch,and what we’d change. The site itself is in good shape: TLS 1.3, HSTS, a strict CSP, every security header we check for, and a security.txt. Mail is the gap. DMARC is p=none and SPF ends in ~all, so nothing tells receiving servers to reject mail that fakes an @heyresearch.xyz address. Once your real mail passes SPF or DKIM, set DMARC to p=reject, and sp=reject for subdomains. p=quarantine is the gentler first step if you want one. Three look-alike domains are registered: • heyreserch.xyz is one letter short, 56 days old, parked, and set up to receive mail. Worth watching. • heyresearch.io sends visitors through another host to a slot gambling site, so it helps if your followers know which domain is yours. • heyresearch.com is listed for sale. Not a risk on its own. On X, several obvious look-alikes of your handle are still unregistered. Claiming a few closes the easiest way to impersonate you. Smaller items: no CAA record, and no MTA-STS, TLS-RPT or BIMI. The full case file is in the video.
1
4
263
Ophiris Recon now starts where attackers start: your name. What’s new: • Mail, DNS, subdomains from certificate logs, TLS and security headers. It tells you who can send mail as you, and these checks cost nothing to run. • Look-alikes. It builds the likely doubles of your domain and your X handle, checks which ones are registered, and opens each site. The ones we’ve found so far were for sale, parked, or redirecting to a gambling site. • Active checks, like looking for an exposed .env or .git, stay off until you prove the domain is yours with one DNS record. We ran seven models on real targets. OrcaCyber GLM 5.3 was the fastest of the accurate ones, so it’s the default now. Live now at ophiris.xyz
3
1
7
442
6 new models on Ophiris. Four from the @OrcaCyberHQ security line, two unrestricted. All six take tool calls, so they run inside agents. Security: · OrcaCyber Ultra 1.0: the top tier. 1M context, 131K output. $5 / $10 · OrcaCyber Zero 1.0: 1M context, 128K output. $3 / $7.50 · OrcaCyber GLM 5.3: same limits as Zero, cheaper output. $3 / $5 · OrcaCyber Hy4: 1M context for $0.80 / $1.60. The budget pick for long runs Unrestricted: · DeepSeek V4.1 Flash Abliterated CSF: writes up to 943K tokens in one reply. $0.315 / $1.05 · SuperGemma4 26B Uncensored: reads images, cheapest of the six. $0.1625 / $0.50 Prices per 1M tokens, input / output. Quick pick: Ultra for the hardest jobs, Hy4 for long cheap loops, DeepSeek CSF for huge outputs, SuperGemma4 for screenshots. ophiris.xyz/models
2
1
1
10
378
$OPHIRIS since launch: - 558 requests - 2.09M tokens processed - 60% served straight from cache That’s real work running through the stack. 87% of those tokens are models reading prompts, context and tool output before they answer. Now the part that matters. AI just made attacks cheap. Recon that used to take a team a week now takes one person with an agent an afternoon. We’ve been saying this since day one. The wave is here, and the teams that got ready early are the ones that come out ahead. $OPHIRIS is building the security stack for that world: Live: security engagements. Set the scope, the rules and the testing window. The recon agent plans, runs its tools and hands back sourced findings with a hash-chained evidence trail and a report you can export. Building: passive infrastructure recon. DNS, SPF, DKIM and DMARC with a spoofing risk read, subdomains from public certificate logs, security headers. Next: a contract audit agent, exploit trace and replay for on-chain incidents, Nmap and Nuclei inside engagements, and purple team mode with an attacker agent and a defender agent working the same scope. Unrestricted models, OSINT, on-chain data and autonomous agents in one encrypted workspace. We haven’t seen anyone else put those together. Hope you prepared, we did. ophiris.xyz/app
1
3
10
393