This page covers the Meko portal, API, and platform. For Installer, MCP server, and agent skills updates, see the [Installer release notes](../../installer/).

### New features

#### Portal

- **Labs** — Open **Labs** in the sidebar and launch [Review a PR with AgentK](../../../examples/labs/). You build context in a datapack, share it with AgentK, read the review, and choose which extracted memories to keep. You can promote one memory for the team. A datapack you create in the lab counts against your plan. Free accounts use the shared AgentK. A dedicated AgentK needs a paid plan.
- **WorkbenchLM tool use** — WorkbenchLM chat can call Meko tools while it answers. A read runs immediately. Saving a memory does too. Deleting a memory, promoting a memory, or updating the datapack waits until you choose **Allow**. **Deny** leaves it unchanged.
- **Standard on AWS Marketplace** — On **Settings > Pricing**, a Free account can upgrade to Standard ($25/month) through AWS Marketplace. After you subscribe, finish account setup in Meko. Until you do, the plan stays on Free. Stripe is marked coming soon.

#### Platform

- **Delete a knowledge-base file** — `knowledgebase_delete_document` deletes one uploaded file. You can make the same delete with an API key.

### Improvements

#### Portal

- **Datapack Analytics** — The **Analytics** tab now also shows context size, a memory breakdown, and recent search activity. It already showed token savings and match quality. **Your datapacks** no longer shows an account-wide efficiency tile, or a context badge of 0 on every card.

#### Platform

- **Knowledge-base upload limits** — Storage rejects a file that is too large or the wrong type. A client that uploaded with a presigned PUT must use the presigned POST.

### Fixed

#### Platform

- **Long conversations** — You can add or delete a message after a conversation's stored history exceeds its size limit. Reading or updating that conversation can still fail while the history is over the limit. Failures no longer interrupt other datapacks.
- **Search embeddings** — Search embeds your query with the same model as the index it searches.
- **Promoted memories** — `memory_search` returns memories you promoted, along with your private memories.
- **Conversation writes** — If a write cannot resolve a datapack, the call fails instead of accepting the message and dropping it.