Zep turns your agent's runs into skills they use at runtime.
͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­͏​  ­
LAUNCH

Hi [First name],
Today we're launching Zep Agent Skills.
Agents in production encounter business problems their prompts and skills don't anticipate. What they learn solving them is lost before similar problems recur.
Zep Agent Skills captures your agent's runs, compiles skills from them, and serves the right skill to your agent at runtime. There's no fine-tuning, and your models stay the same.
The Zep Agent Skills lifecycle. 1 Capture, record each attempt. 2 Compile, draft a skill. 3 Curate, check against evidence. 4 Admit, approve a new version. 5 Serve, deliver to the agent. New learnings feed the next compilation.
Zep's skills make agents more reliable and efficient. On the EnterpriseOps-Gym business process benchmark, Haiku 5.5 with Zep skills improved by 12 points. DeepSeek V4.1 Flash with skills outperformed Opus 5.5 Max without skills.
It's free on every Zep plan until January 4, 2027.
Get started in three steps
  1. Upgrade to the v4 SDK. The Agent Skills methods are only in v4, not the 3.x SDKs.
  1. Send your agent's runs. Record Traces with the SDK, or import the traces you already have in LangSmith or Braintrust.
  1. Serve skills at runtime. Search for skills before each task and record whether the skill worked, through the SDK or the Context MCP Server.
Read the quickstart
I wrote about why we built this, and what our research found, in today's post.
Want to see it work first? Our banking example runs the whole loop on one multi-step task. Haiku fails it on its own, Opus solves it once, and Haiku passes with the skill Zep compiled from that run. Run the example
Daniel
You are receiving this email because you opted in to receive email from the list Product Notifications from Zep AI
Zep AI, 2261 Market Street STE 5686 San Francisco, CA 94114
Unsubscribe