Skip to content

AI Memory in FlyEnv ​

AI Memory gives your AI coding agents a shared place to save preferences, project knowledge and development decisions across sessions. Keep a coding convention available to your next conversation, preserve the reason behind an architecture choice, or let another connected client retrieve what a previous agent saved.

Memories are stored locally in FlyEnv's plugin data directory, with global preferences and project knowledge kept in separate scopes. The plugin provides a 3D memory map, search and editing, client connection setup, backup and restore, and a record of memory tool activity.

Install AI Memory from Settings → Plugin Market, then open its sidebar entry. It has no separate runtime to download or service version to start. Agents access the stored memories through FlyEnv's MCP Server.

AI Memory in FlyEnv showing a project's interactive 3D memory map and memory categories

Keep preferences global and project knowledge separate ​

Use Global for information that applies across projects, such as your preferred language or coding style. Choose a project for its own architecture, setup details, decisions and reusable fixes. A global client connection works across projects; it does not combine their memory collections.

Project memories are associated with a confirmed project root directory. Two unrelated projects with the same folder name stay separate. In the management view, select an existing project or choose a project folder to work with that collection.

The plugin stores memory content and metadata outside your source repository. You can use project instructions to guide an agent's behavior without adding the stored memories themselves to your codebase.

AI Memory scope selector with Global, saved projects and a project path option

Browse a 3D memory map ​

The Memories tab displays the selected collection as an interactive 3D map. Memories are grouped by type: preferences, facts, events, procedures, project states, decisions, observations and notes. Connections indicate shared topics or tags, helping you explore related knowledge.

Drag to rotate, zoom in or out, pause the orbit or reduce motion. Select a category to browse its records, then choose a memory to open its details alongside the map. The list view gives you another way to locate and open a record.

Search by keywords to narrow the collection. Agent search matches titles, summaries, topics and tags; reading the full Markdown content is a separate operation. Give each memory a useful title and summary so it can be found later.

AI Memory preference category with a record list, related memory connections and the selected memory's details

Create, edit and maintain memories ​

Use New memory to add a record yourself, or let a connected agent save it through MCP. A memory includes a type, title, summary, topic, tags, Markdown content and source information.

Open a record to read its content, inspect its source and revision, or edit it. Updates retain revision history and a change reason. Archive removes a memory from normal active results while preserving it for later restoration. Delete permanently removes the record and its history after confirmation.

Each memory body supports up to 16 KiB of UTF-8 text. Keep records focused on one useful fact, decision or procedure so agents can retrieve the relevant knowledge without loading an entire conversation.

AI Memory New memory dialog with title, summary, Markdown content, type, topic, tags and source fields

Connect your coding agent ​

Open Connect an agent and choose an installed client. One-click setup is available for Claude Code, Antigravity CLI, Codex, GitHub Copilot CLI, OpenCode and Kimi.

The connection action prepares the local memory service, writes the client's global MCP connection and installs global memory instructions while preserving existing configuration. You can follow each step's result and retry unfinished steps. Advanced settings also let you install project-specific instructions where needed.

AI Memory Connect an agent page with six supported clients, generic settings and three automatic setup steps for Codex

For other MCP clients, choose Generic settings. Copy the HTTP configuration or use the stdio option when required by your client. Add the memory instructions to the client's supported instruction file or system prompt. The stdio connection requires local Node.js and FlyEnv's bridge.

After setup, reconnect the client or open a new agent session so it loads the connection and instructions. Ask it to save a small sample memory, then retrieve it and report the actual ID and revision. You can also find the record in Memories and inspect the request in Call Records.

Successful setup confirms the configuration was written. Whether an agent saves useful knowledge and retrieves it later depends on its instructions and actual tool calls; automatic saving is not guaranteed.

Five MCP tools for persistent memory ​

AI Memory adds five tools to FlyEnv's MCP server:

  • flyenv_memory_search: find matching memory summaries in a global or project scope.
  • flyenv_memory_get: read a complete memory, including a selected revision.
  • flyenv_memory_store: save a new memory with its source.
  • flyenv_memory_update: revise an existing record with a reason and expected revision.
  • flyenv_memory_forget: permanently delete a record and its history.

Manage tool availability and approval policies in FlyEnv's MCP settings. Keep FlyEnv running with the plugin enabled and the MCP server available while agents use the tools. Different connected clients can access the same stored collection when they use the same scope and project root.

FlyEnv MCP Tools tab showing the AI Memory group, five memory tool switches and approval policies

Back up, restore and reconnect moved projects ​

Open Maintenance → Back up now to create a complete backup folder with integrity checks. To move memories to another computer, copy the entire backup folder and choose Restore from backup in a compatible AI Memory installation.

Restore replaces the current memory collections and first saves a backup of the existing data. Memory backups include global and project records; your project source files and AI client configurations are separate.

When a project has moved, use Project directories in Maintenance to choose its new folder. Reassociation keeps the project identity, memories and revision history while updating its stored path. Available projects can also be moved through Show all projects → Change folder.

If the selected folder already has project memories, FlyEnv asks whether the collections belong to the same project. You can combine them while preserving every record and revision, or keep them separate and choose another directory. Combining does not overwrite or automatically deduplicate memories.

AI Memory Maintenance page with backup, restore, project link checks, project directories and the data folder

Review agent activity ​

The Call Records tab shows memory tool activity and results for the current FlyEnv session. Refresh the list after using an agent, then expand a record to inspect request and response details.

The view keeps up to 200 detailed entries in memory and excludes full memory bodies. These records clear when FlyEnv exits; saved memories persist on disk. Local storage describes where the plugin keeps data: content retrieved through MCP is still passed to your chosen AI client.

AI Memory Call Records tab showing session tool activity, results, durations and a refresh button

Use MCP Server for FlyEnv's broader environment tools, or return to the features overview to explore other plugins.