mirrors ai agent guidance
Top AI Tools

Mirrors AI Agent Guidance: Rules Your Agent Actually Sees

Visit website

Mirrors AI agent guidance delivers plain-language rules to your agent at the right step and measures whether each rule helped. See features and pricing.

mirrors ai agent guidance

Most production AI agents fail quietly: a rule was written, buried in a long prompt, and the agent never saw it when it mattered. Mirrors AI agent guidance fixes that gap by giving teams a guidance system for agents already in production. You write a rule once in plain language, Mirrors delivers it to the agent at the right step of the run, and the outcome is measured with held-out comparisons. Mirrors was shared publicly on Show HN on 16 November 2026, and it is backed by Y Combinator.

What Mirrors Does

Mirrors controls what context your agents are told, and when they are told it. You write a rule once, define where it applies, and preview the past conversations it would have reached before turning it on. At runtime, Mirrors delivers the rule at that step of the run, and you can see which conversations each rule reached, so nothing important hides inside a prompt nobody reads.

Measurement is the other half of the product. Mirrors compares outcomes with and without each rule on real runs, and it surfaces the mistakes your agent keeps making, so you know which rule to write next. A rule that does not help shows up as clearly as one that does, which keeps the rulebook honest as the agent and the product change.

Key Features

Mirrors is built to slot into a stack that already exists. Two calls in your agent are enough: one per user turn to fetch the rules that apply right now, and one to capture the trace when the turn ends. Your app keeps its own model calls, tools and prompts, and Mirrors answers in 200 ms on average, failing open past 750 ms so your agent runs without the rules rather than stalling.

  • Rules written once in plain language, scoped to each agent and workflow
  • Delivery at the right step of the run, with previews against past conversations
  • Held-out comparisons that measure whether each rule actually helped
  • Recurring-mistake detection that shows which rule to write next
  • SDK, API and MCP access in TypeScript and Python, on every plan
  • Works with any framework and any model, from LangChain to Claude Code

Who Mirrors Is For

Mirrors fits teams running AI agents in production who have already felt the prompt problem: rules pile up, each one gets buried among the rest, and nobody knows whether the agent saw the rule or whether it helped. Support agents, sales agents, ops agents and coding assistants all share the same failure mode, and Mirrors is built for exactly that moment.

It also fits teams that need to show their work. When a rule changes behaviour, the held-out comparison is evidence you can take to a review, and the record of which conversations each rule reached is an audit trail for the decisions the agent made. That matters most where a wrong answer costs a customer or a compliance headache.

Mirrors Pricing

Mirrors pricing, checked on 17 Nov 2026, starts with a Hobby plan at $0 a month, which includes 1,000 turns a month shared across unlimited agents, with SDK, API and MCP access, recurring-mistake detection, and rule writing and preview. A turn is one user message plus the agent’s full response, including tool calls and subagents, and retries are never billed. Growth costs $299 a month plus $0.005 per turn, and adds founder onboarding, integration help, 24/7 support, and a usage dashboard with estimated charges.

Enterprise pricing is custom, with volume rates on a usage commitment, plus pricing and billing terms in the contract, support terms agreed in writing, and help planning larger rollouts. Every plan includes unlimited agents with no agent fees, and rules, MCP reads and rule lookups are never charged. Prices may change, so the pricing page is the place to confirm the current figures before choosing a plan.

Getting Started With Mirrors

Starting is free with no credit card required: sign up, install the Mirrors SDK, and send your agent’s conversations so you can write your first rule and preview it against past runs. If the measurement shows the rule helping, Growth is the natural next step for a team in production. Mirrors is worth a look for any team whose agents need guidance they can actually see.

Image: official Mirrors product visual, via mirrors.dev.

KEEP EXPLORING

More tools to discover.

Find your next starting point