Bug bounty
Trace trust boundaries, inspect related code, and draft PoCs for validation.
auth.middleware.tsAbliterated models for authorized offensive security work - trace trust boundaries, draft PoCs, and correlate logs with 1M-token context.
For pentesters, bug bounty hunters, and security analysts
Authorized scope only. Review data handling before upload.
One focused workspace for the evidence-heavy parts of security research.
Trace trust boundaries, inspect related code, and draft PoCs for validation.
auth.middleware.tsCorrelate sanitized logs, compare event patterns, and develop investigation hypotheses.
events.ndjsonExplain obfuscated scripts and inspect payload behavior from code you supply.
payload.jsChoose by context, output limits, and listed throughput.
Trace trust boundaries and work through layered vulnerability hypotheses.
glm-5.3-abliteratedIterate on test ideas, inspect code, and explain unfamiliar behavior.
glm-5.3-flash-abliteratedAnalyze source, draft research scripts, and review proposed fixes.
deepseek-v4.1-flash-abliteratedWhat changes: Abliteration targets refusal behavior; zero refusals are not guaranteed.
What stays: Base-model limitations remain. Validate reasoning, generated code, and findings before acting on them.
Listed speeds are supplied specifications, not independent benchmarks. CyberGym / DEFCON scores are not published.
Start in chat. Build into your research workflow.
Bring repository files, extracted PDFs, and sanitized logs into one conversation. Compare more evidence with a 1M-token context window.
Use the same models through an OpenAI-compatible endpoint in the tools you already work in.
More context. Not unlimited storage.
Fit depends on tokenized input and output needs, not automatic GitHub ingestion.
from openai import OpenAIclient = OpenAI( base_url="https://api.cyrails.com/v1", api_key="YOUR_CYRAILS_API_KEY")result = client.chat.completions.create( model="glm-5.3-abliterated", messages=[ { "role": "user", "content": "Review this code for access-control flaws." } ])print(result.choices[0].message.content)A larger context window does not guarantee exhaustive analysis of every file. Confirm data handling before submitting proprietary vulnerabilities.
Confirm data handling before submitting proprietary vulnerabilities. This module records what is known and what still needs to be verified with the team.
Saved conversations support continuing investigations across sessions.
Confirm prompt logging, retention, and inference-provider handling with the team.
SOC 2 status is unverified. Zero data retention is not established.
Third-party inference providers are used. A zero-retention guarantee is not implied; confirm current terms before uploading sensitive material.
Prepay credits and spend them across chat and API. No monthly subscription.
Buy credits when you need them. One balance covers workspace chat and OpenAI-compatible API calls. Credits are not tokens; model credit rates are shown in your workspace before a request runs.
One balance
Metered at the model rate shown in your workspace. Top up at any time.
Threads, saved conversations, and research notes draw from the same balance.
Every /v1/chat/completions call is metered at the selected model's credit rate.
Committed capacity, invoicing, and access terms agreed before activation.
Credits are not tokens. Model rates apply and are shown in your workspace. No unlimited-token allowance is implied. An API free tier has not been announced. Enterprise capacity and invoicing are agreed before activation.
Ask the curated assistant about refusals, context, data handling, pricing, or API access. Answers come from the product FAQ.
Curated answers. Not a live model.
Curated answers
Start in the workspace, then call the same models from your own tooling. One credit balance across chat and API.
$inspect cyrails --models
GLM 5.31M
GLM 5.3 Flash1M
DeepSeek V4.1 Flash1M
›Bring the question.