Myelinate

Self-improving skills and memory for any AI agent framework, with governance built in.

Myelinate is a gateway your agents call before and after each run. Before a run it returns the skills and memories that apply to the task. After the run it records what happened. In the background it learns: it turns failures, successes and user corrections into reusable skills, improves skills that underperform, and remembers facts and preferences. Every change is reviewed by policy or a human, tested on live traffic, versioned and reversible.

Myelin is the sheath the brain builds around pathways that get used, and that work, so they fire faster and more reliably. Myelinate does the same for agents.

Does it work?

In the house-rules benchmark, an agent learns team conventions it was never told from CI-style feedback alone. On tasks it never trained on, its pass rate went from 0% to 91% after 24 runs.

Why Myelinate

• Works with your framework. Plugs into Microsoft Agent Framework, OpenClaw, LangGraph, CrewAI, the OpenAI Agents SDK, Google ADK, anything that speaks MCP or emits OpenTelemetry traces, and plain REST from any language.
• Learns the right way. Skills are lists of small rules that are edited one at a time, never rewritten wholesale. Each rule tracks whether it helped. Skills only change on reliable evidence (tests, feedback, confident judgments).
• Proves it works. Changed skills are tested against the current version on live traffic before rollout, and a small control group measures how much learning actually improves your success rate.
• Built for enterprises. Tenants, scoped API keys, SSO (Entra ID, Okta, Auth0, Keycloak), approval workflows, an audit log, PII redaction, right to erasure, per-tenant budgets, Prometheus metrics and OpenTelemetry tracing.

How it fits

flowchart LR
    A[Your agent<br/>any framework] -- "1. recall(task)" --> G[Myelinate gateway]
    G -- skills + memories --> A
    A -- "2. record(run, outcome)" --> G
    G -- jobs --> W[Workers: reflect, curate,<br/>evolve, maintain]
    W -- governed changesets --> DB[(PostgreSQL)]
    G --- DB

Where to start

You want to...	Go to
Try it locally in five minutes	Quickstart
Understand skills, episodes, changesets and canaries	Concepts
Connect your agent	Integrations overview
Run it in production	Docker, Kubernetes, Production checklist
Set up SSO and permissions	Security and SSO
Decide what gets auto-approved	Governance
Look up a setting or endpoint	Configuration, REST API