Claude Code, Codex & Cursor Plugin

Persistent memory for AI coding agents

Your agent forgets everything between sessions. wire-memory fixes that. Decisions, corrections, patterns, and preferences, captured automatically and retrieved when relevant. Backed by Wire containers via MCP.

View on GitHub

Free and open source · MIT license · No account required to try

The Problem

Every session starts from zero

Sound familiar?

No continuity

Your agent forgets everything between sessions. Every conversation starts from zero, no matter how much ground you covered yesterday.

Repeated corrections

"Don't mock the database in tests." You said it last week. And the week before. And you'll say it again tomorrow.

Re-explaining context

"Here's how our project works..." again. Every session, you re-teach the agent things it should already know.

No cross-project awareness

Your coding preferences, communication style, and tooling choices don't carry over. Each project is an island.

How It Works

Memory that works in the background

wire-memory captures context as you work and retrieves it before your agent responds. No manual note-taking, no extra prompting.

Automatic logging

Decisions, corrections, and patterns are captured as they happen. No manual note-taking required.

Automatic retrieval

Before responding, your agent searches past context for relevant decisions and corrections. No extra prompting needed.

Scoped by project

Memory is tagged by project and user. Retrieval stays relevant to the work at hand, not polluted by unrelated context.

Portable via MCP

Stored in a Wire container. Access the same memory from Claude Code, Codex, Cursor, or any MCP-compatible tool.

Transcript capture

Optionally upload full session transcripts to your container. Secrets are redacted automatically. Disabled by default.

Getting Started

Set up in 60 seconds

Pick your editor. Connect. Start coding.

1

Install the plugin

Claude Code
# Add the Wire marketplace (one-time)
/plugin marketplace add usewire/wire-plugins

# Install wire-memory
/plugin install wire-memory@wire-plugins
2

Connect your container

/wire-connect

Authenticates via your browser and links a Wire container for storage. No Wire account required. Get a free 7-day trial container instantly.

3

Start coding

That's it. wire-memory automatically captures decisions and corrections as you work, and retrieves relevant context at the start of each session. No configuration, no extra prompting.

Requirements: Node.js ≥ 18 and Claude Code, Codex, or Cursor. wire-memory has zero npm dependencies.

Memory Types

What your agent remembers

Four types of context, captured automatically.

decision

Decisions

Architectural choices and trade-offs

"Use Drizzle ORM over Prisma for Cloudflare Workers compatibility"

correction

Corrections

Feedback when you redirect the agent

"Don't mock the database in integration tests. Use a real test DB"

pattern

Patterns

Coding conventions and established practices

"All API routes use requireAuth() middleware before requirePermission()"

preference

Preferences

Global style and communication preferences

"Keep responses concise. No trailing summaries after code changes."

MCP Tools

Five tools, zero configuration

wire-memory exposes these MCP tools to your agent automatically.

wire_search

Full-text and semantic search across your stored memory

wire_write

Log decisions, corrections, patterns, and preferences

wire_explore

Browse what's stored: types, schemas, entry counts

wire_navigate

From a known entry, reach its siblings, full source, or relationships

wire_delete

Remove outdated or incorrect entries

Built on the Wire SDK

The same primitives any team can ship on

wire-memory is a first-party app on top of @usewire/sdk. The connect flow you ran is a few lines of TypeScript. The same SDK is public, so whatever you are building, an agent, a harness, a plugin, a backend, an internal tool, you can hand it a scoped Wire container the same way wire-memory does.

Three methods

connect, getStatus, disconnect. That is the whole connection lifecycle.

Browser-driven auth

Wire owns the connect screen and authorization. Your code shows a short code, the user picks a container, you get a connection back.

Scoped MCP credentials

connect() resolves with an MCP URL and API key scoped to one container. Hand them straight to your agent's MCP client.

Runs anywhere agents run

Node 18+, Cloudflare Workers, Deno, Bun. Light dependencies, Web Crypto under the hood.

wire-sdk on GitHub @usewire/sdk on npm
TypeScript 
// Same SDK wire-memory uses internally
import { WireClient } from '@usewire/sdk';

const client = new WireClient({ agentId: 'my-agent' });

// Shows a short code, opens the user's browser,
// resolves once they pick a container.
const connection = await client.connect({
  label: 'my-laptop',
});

// Hand to your agent's MCP client:
connection.mcpUrl;
connection.apiKey;
connection.containerId;

Frequently asked questions

Common questions about wire-memory.

What is wire-memory?
wire-memory is a plugin for Claude Code, Codex, and Cursor that gives your AI coding agent persistent memory across sessions. It automatically captures decisions, corrections, coding patterns, and preferences, then retrieves relevant context before each response. Your agent stops repeating mistakes and starts building on past work.
Is wire-memory free?
wire-memory is free and open source under the MIT license. When you connect, you get a free 7-day trial container with no account required. Create a Wire account anytime to keep your memory permanently. Wire containers use a credits-based pricing model. See our pricing page for details.
Which editors does wire-memory support?
wire-memory is a native plugin for Claude Code, Codex, and Cursor. The same usewire/wire-plugins marketplace serves all three. Your memory is stored in a standard Wire container with MCP access, so you can also connect from any other MCP-compatible tool to read and search your stored context.
What data does wire-memory store?
wire-memory stores four types of entries: decisions (architectural choices and trade-offs), corrections (when you redirect the agent's approach), patterns (coding conventions and practices), and preferences (global style and communication preferences). All entries are tagged with project and user context for scoped retrieval.
Where is my data stored?
Your memory is stored in a Wire container, a Cloudflare Durable Object that you own and control. Data is encrypted in transit and at rest. Your API key is stored locally with owner-only permissions. Nothing is shared without your explicit consent.
Do I need a Wire account to try wire-memory?
No. The connect flow creates a free trial container you can test for 7 days with no account required. If you want to keep your memory permanently, create a Wire account and the container transfers to your organization.
How does automatic retrieval work?
wire-memory includes an integrated skill that teaches your agent when to search and write memory. Before responding to your prompts, the agent automatically searches for relevant past decisions, corrections, and patterns. This happens in the background with no extra prompting needed from you.
Can I build my own agent or tool on the same SDK?
Yes. wire-memory is a first-party app on top of @usewire/sdk, the same SDK any team can install. The SDK gives you connect, getStatus, and disconnect. Wire owns the connect screen and authorization, then hands your code a scoped MCP URL and API key for the container the user picked. Use it from an agent, a harness, a CLI, a plugin, or a backend, anywhere you can run TypeScript.
Can my team share a memory container?
Yes. Point wire-memory at a shared container and commit the configuration to your repo. Every team member's agent writes to and reads from the same container. Entries are tagged per-user so retrieval stays relevant to each person while building a shared knowledge base.
What is transcript capture?
Transcript capture is an opt-in feature (Claude Code only) that uploads full session transcripts to your Wire container. Transcripts are stored in native JSONL format, secrets are automatically redacted before upload, and tool result contents are stripped. Enable it via /wire-configure. It is disabled by default.

Give your agent a memory

Install wire-memory and stop repeating yourself. Free and open source.

View on GitHub