How to Give OpenAI Dots Persistent Memory with Memwyre

TL;DR: OpenAI Dots (announced September 2026) are always-on ChatGPT agents powered by GPT-6 Astra. They manage their own cloud computers, browsers, and task queues - but their built-in memory is a consumer-grade silo disconnected from your developer tools. By connecting Memwyre's MCP server to ChatGPT, your Dots gain access to a structured, cross-tool memory vault that syncs with Claude Code, Cursor, OpenCode, Grok, and Muse Code - giving them true institutional memory about your projects.
What Are OpenAI Dots?
Announced at OpenAI DevDay on September 29, 2026, Dots are persistent AI agents that live inside ChatGPT, Slack, and Microsoft Teams. Unlike standard ChatGPT conversations that end when you close the tab, a Dot:
- Runs continuously - it works in the background even when you're offline
- Has its own cloud computer - can browse the web, run code, and manage files
- Manages ongoing tasks - tracks project status, monitors deadlines, and follows up autonomously
- Delegates to subagents - spawns background agents for parallel workstreams
This makes Dots ideal for long-running research, project management, monitoring, and coordination tasks.
The Memory Problem with Dots
Dots have access to ChatGPT's native memory system, but that system was designed for consumer preferences, not developer project knowledge:
The Solution: Memwyre MCP + OpenAI Dots
The Memwyre MCP server (already live at server.memwyre.tech/mcp) provides a structured memory API that any MCP-compatible client - including ChatGPT's connected app ecosystem - can use.
Once connected, your Dot can:
- Search past decisions made in any of your connected tools
- Save new insights from Dot sessions back to your shared vault
- Cross-reference context across your entire AI developer stack
- Share knowledge with other Dots connected to the same vault
Setup: Connect Memwyre to ChatGPT Dots
Step 1: Generate Your API Key
- Go to app.memwyre.tech/settings
- Click API Keys → Generate New Key
- Copy the key (format:
bv_sk_...)
Step 2: Add Memwyre as a Connected App
In ChatGPT:
- Enable Developer Mode in your ChatGPT settings (required for custom MCP connections)
- Click your profile → Settings
- Navigate to Connected Apps (or Integrations)
- Click + Add MCP Server
- Enter:
- Name: Memwyre Memory Vault
- URL:
https://server.memwyre.tech/mcp - Auth Type: Bearer Token
- Token: Your
bv_sk_...API key
Production note: For multi-user or enterprise deployments, OpenAI requires OAuth 2.1 with PKCE instead of static Bearer tokens. The static token approach works for personal use. Contact support@memwyre.tech for OAuth integration guidance.
Step 3: Enable for Your Dot
- Open your Dot's configuration
- Under Capabilities, find and enable Memwyre Memory Vault
- Optionally, add a standing instruction to your Dot:
"Before starting any project task, search Memwyre for existing decisions and context about that project. After completing research or analysis, save a summary to Memwyre for future reference."
⚠️ Read-Only Proactive Mode
When your Dot operates autonomously in the background (proactive mode), OpenAI enforces a strict read-only restriction:
- Read tools work normally:
search_memwyre,list_memories,get_inbox - Write tools are staged:
save_memory,update_memory,delete_memorywill be queued and require your explicit approval before executing
This is an OpenAI platform safety constraint and cannot be bypassed with Custom Rules.
Available Memwyre Tools for Dots
| Tool | When to Use |
|---|---|
search_memwyre |
Look up past decisions, patterns, and context |
save_memory |
Store new findings, decisions, and notes |
list_memories |
Browse recent memories (useful for orientation) |
get_inbox |
Review memories pending your approval |
update_memory |
Correct or update an existing memory |
delete_memory |
Remove outdated or incorrect memories |
search_by_date |
Find what was decided in a specific time period |
generate_prompt |
Generate a context-enriched prompt for a task |
📊 Evaluation Backing: Wondering how cross-tool memory performs at scale? Read our LoCoMo-10 Benchmark Report showing a 70.5% accuracy rate and 88% token reduction, or explore our research on Vector Database vs. Agent Memory.
Real-World Use Cases
1. Project Kickoff Researcher Dot
Configure a Dot to run at 9 AM every weekday:
"Search Memwyre for all memories tagged with today's active project. Prepare a morning briefing with: current status, open decisions, next actions."
The Dot surfaces all context from your previous sessions across all tools - no manual catch-up needed.
2. Architecture Review Dot
Before implementing a new feature:
"Search Memwyre for conventions related to [feature area]. Check for existing patterns, anti-patterns, and agreed standards. Report any conflicts with the proposed implementation."
3. Cross-Team Knowledge Bridge
When a new developer joins a project, their Dot can onboard them automatically:
"Search Memwyre for project conventions, setup guides, and architectural decisions. Summarize the most important things for a new developer to know."
How This Compares to Native ChatGPT Memory
| Capability | ChatGPT Native Memory | Memwyre Vault |
|---|---|---|
| Cross-tool sync (Cursor, Claude, Grok) | ❌ | ✅ |
| Structured code knowledge | ❌ | ✅ |
| Team/multi-user sharing | ❌ | ✅ |
| API-queryable | ❌ | ✅ |
| Inbox curation & approval | ❌ | ✅ |
| Tagged & searchable | Limited | ✅ |
| Context window efficiency | Uncontrolled | ✅ Semantic top-K retrieval |
Privacy & Security
All communication between ChatGPT and Memwyre happens over HTTPS (TLS 1.3). Your API key is sent as a Bearer token per request - it is never cached by the MCP server. Memory data is encrypted at rest (AES-256). See memwyre.tech/privacy-policy.
Related Reading & Agent Plugins
- Best AI Agent Memory Systems in 2026: Review of Memwyre, Mem0, Letta, and Zep.
- OpenCode Memory Plugin: High-performance persistent memory for the open agent runtime.
- LoCoMo Benchmark Report: Long-term context recall evaluation.
- How to Vibe Code Without Limits: Prevent API quota exhaustion.
- Muse Code Persistent Memory: Lifecycle hooks for Meta Muse Code.
Get Started
Connect your Dots to Memwyre in under 2 minutes:
- Get an API key: app.memwyre.tech/settings
- Connect MCP: ChatGPT → Settings → Connected Apps →
https://server.memwyre.tech/mcp - Configure your Dot to use Memwyre as its project memory backend
