Announcing our public launch on Y CombinatorRead more (opens in a new tab)
Backed byYCombinator

Give agents the right context,
exactly when they need it.

Infrastructure to control what your agents are told, and when.

Start free. No credit card required.See how it works
works with any framework and any model
  • LangChain
  • LangGraph
  • LlamaIndex
  • CrewAI
  • AutoGen
  • Pydantic AI
  • smolagents
  • Google ADK
  • Hermes
  • AI SDK
  • Mastra
  • OpenAI
  • Anthropic
  • MCP
  • Claude Code
  • Codex
  • Cursor
  • Windsurf

The rule was there. Your agent never saw it.

It was buried in a long prompt, or sitting in a skill the agent never opened.

You can write the rule. You can’t control whether your agent sees it when it matters.

How it works.
From a rule written once to a measured result.

From a rule written once to a measured resultTurns, outcomes, and feedback reveal the mistakes your agent repeats. You write a rule and preview where it applies. Mirrors delivers it when the situation matches, then compares outcomes with held-out conversations using the outcomes your team tracks.HOLDOUT / WITHOUT THE RULESupportworkflowRefundroutingEnterpriseaccountsTriggersscopeTurnsignalsBillingworkflowScopedruleReviewedrules02 / RULES THAT FIT01 / MIRRORS ENGINE03 / YOUR AGENT right rule · right stepoutcome report04 / LEARN FROM THE OUTCOMEYOUR AGENT’S EXPERIENCEINPUT// a recorded turn03 update_policy coverage: comprehensive// customer feedback“Check the license first.”// a successful resolutionverified · resolved · confirmedTurns · Outcomes · FeedbackEXAMPLEinformed byroutes rulesrules ↔outcomesholdout outcomes →Write the rule.Use it next time.
01

Write a rule once.

Say what your agent should do, in plain language.

02

Define where it applies.

Preview the past conversations it would have reached.

03

Mirrors delivers it.

Your agent gets the rule at that step of the run, and you see which conversations it reached.

04

Measure whether it helped.

Compare outcomes with and without the rule, and see which mistakes still repeat.

Two calls in your agent. The rest of your stack stays as it is.

200 ms on average
Fails open: past 750 ms, your agent runs without the rules.
Any framework, any model
Your app keeps its own model calls, tools and prompts.
SDK, API and MCP
TypeScript and Python, on every plan.
Read the quickstart
agent.tsTypeScript
import { Mirrors } from "mirrors-sdk";

const mirrors = new Mirrors({ apiUrl, apiKey, agentId });

// Once per user turn: the rules that apply right now.
const { context } = await mirrors.getContext({ trace });
const system = context ? `${prompt}\n\n${context}` : prompt;

// Your model call, tools and streaming stay as they are.
mirrors.captureTrace({ trace, final: true });
npm install mirrors-sdk

Write a rule once. Decide where your agent sees it.

See the mistakes your agent keeps making.

So you know which rule to write next.

Know where a rule will land.

Preview every conversation it would reach before you turn it on.

Every rule in one place.

Scoped to each agent and workflow. Update one, not your whole prompt.

The right rule, at the right step.

Mirrors delivers it only where you said it applies.

Know which rules help.

Compare outcomes with and without each rule.

Rules your agent sees when they matter.
Before and after Mirrors.

TODAY WITH MIRRORS
01Rules get buried in a long prompt, or sit in a skill the agent never opens.
↗Each rule reaches your agent at the step where it applies.
02Every rule rides along on every request.
↗Rules are delivered only where you said they apply.
03No one knows whether the agent saw the rule.
↗You see which conversations each rule reached.
04No one knows which rules help.
↗Every rule is measured.

63,052 agent turns. And counting.

Try it free on your agent

Pricing

Hobby

$0/month

1,000 turns a month, free

Start free

INCLUDES

  • Unlimited agents
  • SDK, API and MCP access
  • Find recurring mistakes
  • Write and preview rules

Enterprise

Custom

Volume rates on a usage commitment

Chat with us

INCLUDES

  • Unlimited agents
  • Pricing and billing terms in your contract
  • Support terms agreed in writing
  • Help planning larger rollouts
See everything in each planHide plan details
What each Mirrors plan includes
HobbyGrowthMost teamsEnterprise
PRICE
Base fee$0Custom
Turns1,000 a month, free, shared across all your agentsVolume rates on a commitment
AgentsUnlimited, no agent feesUnlimited, no agent fees
RulesNo chargeNo charge
MCP reads and rule lookupsNo chargeNo charge
Rate guaranteeDoes not applySet in your contract
BILLING
BillingNoneSet in your contract
Turns billedDoes not applyEvery turn counted
Usage dashboard and estimated chargesDoes not applyIncluded
Invoices and payment methodsDoes not applyIncluded
CancellationDoes not applySet in your contract
PRODUCT
Find recurring mistakesIncludedIncluded
Write and preview rulesIncludedIncluded
Publish rules and deliver them at runtimeIncludedIncluded
SDK, API and MCP accessIncludedIncluded
SUPPORT
Founder onboardingNot includedIncluded
Integration helpNot includedIncluded
SupportEmailTerms agreed in writing
Larger implementation workNot includedScoped and quoted separately

A turn is one user message and your agent’s full response, tool calls and subagents included. Retries are never billed.

Full pricing details

FAQ

What does Mirrors do?

Mirrors is the guidance system for production AI agents. You write a rule once and define where it applies. Mirrors delivers it to your agent at that step of the run, shows which conversations it reached, and measures whether it helped. It also surfaces the mistakes your agent keeps making, so you know which rule to write next.

Why not put the rule in my system prompt or a skill?

A rule in the system prompt rides along on every request, so rules pile up and each one gets buried among the rest. A skill is read only when the model decides to open it, so the rule can be missed. Mirrors delivers each rule where you said it applies, keeps every rule in one place, scoped to the right accounts and workflows, and shows whether it helped.

Does Mirrors guarantee my agent follows the rule?

No. Mirrors makes sure the rule reaches your agent where you said it applies, and shows you the conversations it reached. Whether the model then follows it is what the measurement is for: you compare outcomes with and without the rule on real runs.

How is Mirrors different from evals and observability tools?

Observability shows what your agent did. Evals check it against test cases before you ship. Mirrors controls what your agent is told in production: it delivers your rules where they apply, shows which conversations each one reached, and measures each one on real runs. It works alongside both.

How do I connect my agent?

Install the Mirrors SDK, send your agent’s conversations, and request the rules that apply once per user turn. It works with any agent framework and any model, and your app keeps control of its model calls and tools.

Does Mirrors add latency?

Mirrors answers in 200 ms on average, once per user turn. It fails open: if a response takes longer than 750 ms, your agent continues without the rules.

Can I check a rule before it goes live?

Yes. Mirrors previews each rule against your stored conversations, so you see every one it would have reached, including any you didn’t expect, before you turn it on.

How much does Mirrors cost?

Every plan includes unlimited agents. Hobby is free with 1,000 turns a month shared across your agents, and you can sign up without talking to anyone. Growth is $299 a month plus $0.005 per turn. Enterprise is custom pricing for high volume.

What your agents are told becomes how your company runs.

Start free with 1,000 turns, or book a demo.

Start free