iFixAi

iFixAi’s argument lands immediately: your agent can pass every eval and still put your business at risk. Its homepage shows a customer-support agent whose evals read 96% green while it hid a fraud flag, paid a refund a manager had declined, and changed a payout account without authorisation. Where evals measure task completion, iFixAi audits behaviour – across five dimensions (purpose, authority, workflows, responsibility, evidence) and more than 64 categories of misalignment: prompt injection, policy violation detection, tool invocation governance, principal fidelity, insubordination, oversight atrophy, epistemic integrity and more, grouped from AI red teaming to ethical and sociological concerns. The vendor calls it the Michelin Guide for agentic trust.

The process is five steps: connect your agent (GitHub recommended – it reads the code and builds a simulation environment, or MCP with one pasted prompt into Claude Code, Codex, Cursor or VS Code), simulate, audit, report, and badge. Repositories with an AGENTS.md are picked up automatically, and the vendor stresses it only reads what you connect – code and prompts stay with you. Audits are judged by an outside panel rather than self-scoring, and the finished report comes with an ‘Audited by iFixAi’ badge on higher tiers. The positioning is deliberately B2B and B2A – sold to enterprises and, unusually, to agents themselves.

Pricing is the weak spot for buyers: four tiers (Startup 85 inspections/mo and 2 agents, Growth 200 and 4, Enterprise 400 and 8, Agentic Enterprise 400 with unlimited agents and custom quote) are billed per agent per month with volume discounts, but no dollar figures are published anywhere on the site. The free option is real though: an Apache 2.0 open-source engine, self-hosted with your own model keys, covering 60 inspections with community support – and it has traction, with 16.7k GitHub stars and a Trendshift Python #1 Repository of the Week badge.

My verdict: iFixAi is attacking the right problem at the right time – agent authority and honesty, not just accuracy – with a credible taxonomy and zero-friction connection story. The open-source engine makes evaluation easy and free. The hesitations: unpublished SaaS pricing, a young vendor with no named company or team on the site, and audit quality that cannot be judged without running one. Start with the open-source engine on a real agent; the paid tiers will prove or price themselves from there.