Supercharge your SDLC with AI agents. They plan, design, code, test, secure, deploy and maintain. You stop prompting. The work just lands.
Type a request. Watch agents do the work. Real files, real tests, real cost. Same flow that runs in your terminal.
Looping preview. Your terminal runs the same chain.
A single chatbot writes a function then stops. You become the project manager. VxAI fixes that.
No spec. No tests. No security check. You spend the rest of the day prompting it into shape.
Your spec PDF, your conventions, your domain notes. Every new session starts from zero.
Most CLIs lock you to one model. So you either overpay on cheap work, or underperform on hard work.
Most tools hide tokens behind a one-line summary. You cannot budget what you cannot see.
Master breaks your request into phases and runs each specialist with its own workflow. No skipped steps, no "I forgot the tests".
Claude Opus for design. GPT-4.1 for code. Haiku for tests. Both Anthropic and OpenAI in one tool. Pick per agent.
Rules, pinned docs and notes per project. PDF, Excel, Word and CSV extracted to text. Every agent reads them every turn.
Live tokens and dollars per query in your terminal. Prompt caching on by default. No invoice surprises.
Agents read, write, grep, and run shell commands inside your working directory. Real diffs, real test runs, real CI files.
Your keys, code and history stay on your machine. Tool calls sandboxed to the working directory. Only prompts go to the model.
Nothing is auto-scanned. You pin the docs, you write the rules, you save the facts.
VxAI carries that curated context into every agent run, so your security policies and conventions
never drift.
Read the detailed Context study →
Hard constraints applied on every turn. "Never log secrets or PII." "Use Prisma, never raw SQL." Written once, enforced on every run.
Nothing auto-ingested. You pin a PDF, Excel, Word or CSV and VxAI extracts the text once. The team reads only what you approve. Nothing silent.
Project truths the team carries forward. "Prod DB is read-only." "All endpoints require JWT auth." Saved once, never re-explained.
| 01. | stripe-checkout-spec.pdf | 84 KB | 14 pages | |
| 02. | XLSX | pricing-tiers.xlsx | 12 KB | 3 sheets |
| 03. | MD | ARCHITECTURE.md | 6 KB | live read |
The agent you talk to. Reads your request, splits it into phases, and routes each one to the right specialist. Then puts the answer together.
Turns a fuzzy ask into a real spec. User stories, acceptance criteria, scope. Saves it to
docs/planning/.
Architecture, file tree, APIs, data shapes, trade-offs. The blueprint the developer follows.
Reads files before editing them. Writes real code, no TODO stubs. Runs typecheck before saying done.
Picks the test runner, writes happy-path and edge cases, and iterates until everything passes.
OWASP scan, secret leaks, dep audit, compliance cues for SOC 2 and friends. Returns severity, file, line and fix.
GitHub Actions, Dockerfiles, release scripts. Pinned versions. No hardcoded secrets. Rollback plan included.
Bug triage with root cause and smallest fix. Or tech-debt audits, ranked by priority and effort.
Open /settings, drop in an Anthropic or OpenAI key (or both), pick a model per
agent. Saved locally at ~/.vxai/settings.txt.
"Add Stripe checkout." "Audit for SOC 2." "Fix the failing test." Master decides if it's a chat or a chain.
Tool calls visible live. Real files, real tests, real cost on screen. Press Esc to cancel anytime.
Pick any model per agent. Switch any time. Forget your key for a model? VxAI tells you in plain English instead of crashing.
All current Claude models with prompt caching on by default. One key, every Claude-routed agent.
GPT-4.1, GPT-4o, o4-mini and their mini variants. Tool calls translated under the hood. Same agent workflow either way.
Local install. Your keys, your machine, your code. Beta
$npm install -g @venex/vxai
$vxai
launch
No credit card · Bring your own Anthropic / OpenAI key · Settings stay local in ~/.vxai/