mirrorsChat with a founder
Y Combinator logoBacked by Y Combinator

The right context.
At the right time.

Mirrors learns from your agent’s successes and failures to route the right context at the right moment, so the same mistakes stop reaching your customers.

Chat with a founder
built for your agent’s context
  • LangChain
  • LangGraph
  • LlamaIndex
  • CrewAI
  • AutoGen
  • Pydantic AI
  • smolagents
  • Google ADK
  • Hermes
  • AI SDK
  • Mastra
  • OpenAI
  • Anthropic
  • MCP
  • Claude Code
  • Codex
  • Cursor
  • Windsurf

Your agent has the instructions. It needs the right ones in the moment.

Every conversation teaches you something about your agent. Where it gets stuck. What gets it back on track. Which instructions lead to a good outcome. Too often, those lessons stay buried in the traces.

Mirrors turns those lessons into context you can route. Decide what your agent should see, preview the situations where it belongs, and deliver it when those situations come up again. Every change has a scope and an outcome you can measure.

Learn from every session. Improve the next one.
Context that changes with the conversation.

From experience to context at the right timeSessions, outcomes, and feedback reveal patterns in agent behavior. You review the guidance and preview where it belongs. Mirrors routes context when the situation matches, then compares outcomes with held-out sessions to measure whether the change helped.HOLDOUT / WITHOUT NEW CONTEXTGuidancecontextRulesroutingToolscontextTriggersscopeSessionsignalsResultscontextPromptguidanceReviewedlessons02 / CONTEXT THAT FITS01 / MIRRORS ENGINE03 / YOUR AGENT right context · right timeoutcome report04 / LEARN FROM THE OUTCOMEYOUR AGENT’S EXPERIENCEINPUT// a recorded session03 update_policy coverage: comprehensive// a customer correction“Check the license first.”// a successful resolutionverified · resolved · confirmedSessions · Outcomes · FeedbackEXAMPLElearns fromroutes contextcontext ↔outcomesholdout outcomes →Learn the lesson.Use it next time.
01

Learn what works. Learn what doesn’t.

Every successful resolution, retry loop, and customer correction holds a lesson. Mirrors finds the patterns in the sessions your agent already ran.

02

Give each lesson a moment to matter.

Write the context your agent needs. Preview which situations should receive it, including unexpected matches, before putting it live.

03

Deliver the right context, right on time.

As the conversation unfolds, Mirrors routes relevant guidance when its conditions match. Your agent gets the lesson where it can use it.

04

Stop paying for the same mistake twice.

Compare outcomes with and without the added context. Keep what helps, refine what doesn’t, and turn yesterday’s failure into a better next conversation.

Learn from experience. Act in the moment. Make the next conversation better.

Every session has something to teach.

Find what worked, where things broke, and what context was missing.

Image

The right context. Exactly when it matters.

Route guidance as the conversation unfolds, when its conditions match.

Know where your context will land.

Preview matching, missed, and unexpected sessions before activation.

Image

Give your agent a better next move.

Edit guidance, tool descriptions, schemas, and model-facing results.

Image

Fewer repeat mistakes. Happier customers.

Measure which context changes improve the outcomes customers care about.

Your agent’s context shouldn’t stand still.
It should respond to what’s happening.

TODAY WITH MIRRORS
01Successful sessions disappear into the logs.
↗What worked becomes evidence for the guidance you give next.
02Every failure adds another instruction to the global prompt.
↗Each lesson reaches the situations where it belongs.
03Your agent sees the same instructions as the conversation changes.
↗Relevant context arrives when the moment calls for it.
04Customers keep explaining the same mistake.
↗Measure which context changes reduce recurring failures.

Fewer repeat mistakes.

A lesson only matters if the next conversation goes better. Compare outcomes with and without the added context. See what helps, refine what doesn’t, and give your customers fewer reasons to complain.

See what that looks like for your agent

Start with
one workflow.

Bring your traces and a workflow you want to improve. We’ll work through the context it needs and discuss pricing for your use case.

Let’s talk / pricing
01

Find the moments where better context could help.

02

Turn one recurring mistake into a lesson your agent uses.

Chat with a founder

FAQ
Common questions.

What do you mean by dynamic context?

Giving your agent the guidance it needs for the situation it is in. Mirrors routes context based on what is happening in the session, so a lesson from an earlier conversation can help at the moment it matters.

How does Mirrors learn from successes and failures?

Mirrors uses recorded sessions, outcomes, and feedback to find recurring patterns and identify where context could help. Successful sessions show what worked; failed sessions show what needs attention. You write and review the guidance that gets routed.

What context can I change?

You can append guidance, edit a tool description, tighten a tool’s input schema, or rewrite the tool-result content shown to the model. Tool descriptions and schemas are set at session start; guidance and result edits can apply during the conversation.

How do I know the context reaches the right sessions?

Preview the routing conditions against stored traces before activation. Review matching sessions, missed examples, and unexpected matches. You can see where a change would apply before your customers encounter it.

Does routing context slow down my agent?

Routing decisions run locally, without a network request or model inference to decide what applies. Configuration refresh and telemetry happen separately. If no valid configuration is available, your agent continues without new context edits.

What do I need to get started?

Bring exported traces from a workflow you want to improve, along with any outcome labels or customer feedback you already collect. We’ll help identify a recurring pattern, review the context it needs, and preview where that context would go.

How do I know customers are having a better experience?

Choose an outcome you care about, such as successful resolution, and compare sessions that received the context with stable holdout groups. Sample counts and confidence intervals help you judge the result. Revise, expire, or roll back changes as you learn; rollback stops new applications without erasing context already delivered.

The right context can change the next outcome. Put it where it matters.

Book a call. Turn a lesson from your traces into context your agent can use.

Chat with a founder