ReflexioAI
Reflexio is a learning platform and open-source harness for AI agents. It turns user corrections, failed paths, and successful outcomes from real interactions into visible, testable, reversible behavioral improvements that agents can retrieve and reuse.
At a Glance
- Developers and AI application teams
- Enterprise teams operating production AI agents
- Customer support, sales, data analysis, and recruiting organizations
- Teams requiring self-hosting, data residency, or bring-your-own-cloud deployment
- +1 more
AI Tools by ReflexioAI
(1)
Reflexio
AI Agent Self Improvement Harness
Discussions
No discussions yet
Be the first to start a discussion about ReflexioAI
Latest News
Reflexio published “Everyone Is Building Self-Improving Agents. Almost Nobody Means the Same Thing.”
Reflexio published “Your Agent Dreams. It Still Can't Tell You If It Got Better.”, comparing agent-memory consolidation systems and measurement of improvement.
Reflexio published the Creao case study “How Creao Turned Agent Exploration Into Repeatable Execution.”
Reflexio published “Your Agent Remembers Everything and Learns Nothing,” explaining its distinction between memory and behavioral learning.
Products & Services
Managed learning platform for AI agents: publishes interaction evidence, extracts and aggregates profiles and playbooks, retrieves relevant context, and evaluates whether learnings helped or hurt an agent response.
Open-source locally hosted deployment that provides local memory, online playbook learning, retrieval-backed behavior improvement, configurable models/extractors/evaluation, and local storage.
Python client and data models for searching learned context and publishing interactions, including user profiles, user playbooks, agent playbooks, sessions, and evaluation signals.
HTTP API for search and publish_interaction operations, allowing agents to connect to hosted Reflexio without using the Python SDK.
Market Position
Reflexio positions itself as a behavioral-learning and self-improvement layer rather than a conventional memory store. It distinguishes storing what a user said from learning how an agent should behave differently, and emphasizes evaluation, approval, reversibility, playbook reuse, and cross-user transfer. The competitive landscape includes memory products such as mem0, coding-agent memory and dreaming systems such as Claude Code, Codex, Gemini CLI, and OpenClaw, prompt/context optimizers, and weight-level fine-tuning or reinforcement-learning systems; Reflexio's differentiator is a hosted or open-source retrieve-and-publish loop that learns from production interactions without retraining model weights.
Leadership
Founders
Yi Lu
Co-founder and CEO. Before Reflexio, Yi Lu was a tech lead at Meta and an adjunct professor at the University of Washington teaching machine learning and business applications; LinkedIn lists the University of Virginia in his education history.
Guangyu Yang
Co-founder and CTO. Previously a senior machine-learning engineer at Meta and TikTok; LinkedIn describes him as a PhD graduate of Johns Hopkins University.
Executive Team
Yi Lu
Co-founder and CEO
Former Meta tech lead and University of Washington adjunct professor teaching machine learning and business applications.
Guangyu Yang
Co-founder and CTO
Former senior machine-learning engineer at Meta and TikTok; Johns Hopkins PhD.
Founding Story
The founders started Reflexio after observing that AI agents often do not improve with use: an agent may fail a task, receive a correction, and still repeat the same mistake across sessions and users. Their initial vision was to close the production learning loop by observing live traces, extracting lessons from successes, failures, and corrections, and feeding reusable behavioral guidance back into future agent runs without retraining model weights.
Business Model
Revenue Model
Subscription SaaS for the managed hosted platform, with a custom-priced bring-your-own-cloud/self-hosted enterprise option. The open-source local deployment is available under Apache License 2.0.
Pricing Tiers
100 learnings generated per month, 1,000 search requests per month, data retained while the account is active, 30-day Pro trial, and community support.
10,000 learnings generated per month, 100,000 search requests per month, data retained while the account is active, and priority support.
Self-host in the customer's AWS, GCP, or Azure account; data stays in the customer's infrastructure; bring your own LLM and storage; configurable retention; dedicated support and onboarding.
Target Markets
- Developers and AI application teams
- Enterprise teams operating production AI agents
- Customer support, sales, data analysis, and recruiting organizations
- Teams requiring self-hosting, data residency, or bring-your-own-cloud deployment
- Open-source developers building self-improving agents
- Customer-support agents that should learn from corrections and avoid repeated mistakes
- Coding agents that should retain repository-specific workflows, verification steps, and fixes
- Sales assistants
- Data-analysis agents
- Recruiting assistants
- Domain-specific enterprise agents that need reusable procedures and user preferences
- Creao, which used Reflexio to turn successful agent behavior into reusable guidance and reported 75% fewer median model calls, 36% fewer task failures, and 47% positive impact from retrieved learnings in its evaluation.