from nepsys import verify_sync result = verify_sync(response, mode="deep") H = (1 - C_adj) * W(t) * S(v) * U(s) / A(s) assert result.passed # every claim verified providers: 6 | catch_rate: 0.9957 # Trust but Verify — always pipeline = Pipeline([ ("verify", HallucinationFilter()), ("route", ProviderRouter()), ]) for claim in response.claims: if claim.h_norm > THRESHOLD: regenerate(claim)

Nepsys Code

Ship AI‑Generated Code You Can Trust

The only coding agent that verifies every output through cross-vendor hallucination detection before you see it. 6 LLM providers. Zero blind faith.

Get the Code See How It Works

scroll ↓

Trust but Verify

Every response is scored before you see it

Nepsys Code sends every AI output through a second, independent vendor for cross-verification. Claims are extracted, scored via the SHI formula, and flagged if the math says the AI is guessing.

nepsys-code

nepsys> explain how to handle rate limiting in express

To handle rate limiting in Express, use the `express-rate-limit`

middleware. Install via `npm install express-rate-limit`, then:

...

── SHI Report ──

Authority: 0.92 │ Claims: 7 │ Mean H: 0.18

5C / 1t / 1v / 0D

─────────────

nepsys> /review

Review mode ON — file changes require approval

nepsys> add the rate limiter to my app

── Proposed changes (1) ──

[1] edit_file

path: src/server.ts

┌─

│ - import express from 'express';

│ + import express from 'express';

│ + import rateLimit from 'express-rate-limit';

└─

Approve? [y/N] y

Feature Set

Everything you need. Two things nobody else has.

15 of 15 Claude Code features matched — plus external hallucination verification and multi-vendor resilience.

🛡

SHI Hallucination Verification

Every response scored through cross-vendor verification. Claims are extracted, checked by a different LLM, and gated by the SHI formula. No other coding agent does this.

🔄

6-Provider Fallback

Claude, GPT, Grok, Gemini, DeepSeek, Perplexity. If one vendor goes down, the next picks up. Your workflow never stops.

⚡

Plan Mode

Break complex goals into steps. Review the plan. Approve, then execute. Human-in-the-loop by default.

👁

Review Mode

See a colored diff of every proposed file change before it executes. Approve or reject with one keystroke.

🔧

Skills & Slash Commands

Reusable prompt templates in SKILL.md files. Type /investigate, /plan, /deploy — your workflows, codified.

🪝

Lifecycle Hooks

PreToolUse can block destructive operations. PostToolUse for audit logging. SessionStart for environment setup.

🧠

Sub-Agent Delegation

Spawn focused sub-agents for codebase exploration, file audits, and research. Results flow back to the main conversation.

🔌

MCP Server Support

Connect any Model Context Protocol tool server. Filesystem, GitHub, Postgres — the MCP ecosystem is your toolbox.

🌲

Worktree Isolation

Run destructive operations in a temporary git worktree. Merge back, keep for review, or discard — your main branch stays safe.

💰

Cost Control

Per-model pricing. Session, daily, and weekly budget caps. Automatic tier downgrade when approaching limits.

📦

VS Code Extension

Ctrl+Shift+F opens Nepsys Code in your editor. Same verification, same tools, same trust guarantees.

🔑

Session Resume & Memory

Pick up where you left off. Per-project memory auto-learns your patterns, decisions, and preferences.