Overview

Uteke provides a single long‑term memory store that any AI agent speaking MCP or plain HTTP can read from and write to. The service runs as a cloud endpoint at https://uteke.app/ and also offers an open‑source binary you can host yourself. A free open‑beta account gives you 500 stored memories and 150 recall requests per day, with no credit‑card required.

The core idea is that every agent you connect, whether a chat assistant on your phone, a coding agent in a terminal, or a custom tool you build, writes its observations to the same memory space. Each entry records who created it, when, and from which device, so the knowledge persists even when you switch models or devices. You interact with the store through three simple steps: add the Uteke URL to your agent's connector or MCP settings, sign in with a CodeCora ID (or use an API key), and then issue “remember” or “recall” commands as part of normal conversation. No code changes are required for supported agents, and step‑by‑step guides exist for Claude, ChatGPT, Claude Code and Cursor.

For developers, the service exposes a REST API at https://uteke.app/api/v2 and an MCP endpoint at https://uteke.app/mcp. A typical “remember” request posts JSON containing a content field and optional tags; the system indexes the text on arrival using both keyword and vector methods, then stores it in a PostgreSQL backend. A “recall” request posts a query field and receives a fused list of results ordered by relevance, there is no separate model‑based ranking step. You can list, read, edit, or delete any memory, and every change is logged with a history entry. Export is available as a single NDJSON stream that includes all memories, documents and tags, allowing you to back up or run the same engine locally.

Key features

One shared memory space for any MCP or HTTP‑capable AI agent
Search‑based recall (keyword + vector) without a model reading the data
Editable entries with full change history and author metadata
Export to NDJSON at any time, no lock‑in or export fees
Open‑source engine you can self‑host, matching the cloud version
Free tier includes 500 memories and 150 daily recalls

Details

Pricing: Free

Suggest an edit

Sign in to suggest an edit.

Pros and cons

Pros

Works with a wide range of agents, including popular chat and coding assistants
No AI model runs on recall, keeping the process fast and privacy‑focused
Secrets such as API keys are masked before storage
Export and self‑hosting options give full data control

Cons

Free tier limits to 500 memories and 150 recalls per day, with paid plans not yet listed
All memories are stored in a single Singapore region, limiting geographic redundancy
Documentation mentions “45 tools” but does not list them, so compatibility may need verification.