Custom Agent — Your Own Agent, One Shared Brain
Frontier Agent — Their Agent, Your Shared Brain
OpenClaw shown as the example custom agent · shared brain over MCP · Tailscale-protected · model-agnostic · you own the data and pick the models.
A leading AI platform runs the agent · it reads and writes your shared brain over MCP · your memory stays yours if you switch providers.
Github · Repositories · Sync Layer
Durable sync between local, repos, and VPS
Durable sync between local, repos, and your server
Agent Brain
private
Source of truth for
identity, memory, rules,
memory, rules, and the
tools, and project context.
project context every agent reads.
Agent Portal
public-facing
Dashboard for brain, tasks,
projects, people + share links.
Deploys via Netlify.
Skills Master
optional
Portable skills and MCP
server configs, kept in sync
across environments.
AI Model Layer · Priority Routing
AI Model Layer · Built Into Your Agent
Model-agnostic. No single-vendor lock-in.
Uses your frontier agent's own models.
1
•
OpenAI Codex
Claude
subscription · primary coding + reasoning
in Claude Duet
2
•
Anthropic Claude
ChatGPT
subscription · long-context reasoning, drafting
in ChatGPT Dots
3
•
Paid OpenRouter models
Grok or Muse
fallback layer · routes across multiple vendors on demand
in Grok Bot or Muse: whichever platform you choose
Agent Configuration Files
Read at startup — shapes every session
IDENTITY.md
USER.md
MEMORY.md
TOOLS.md
HEARTBEAT.md
AGENTS.md
project files
reference files
Identity tells the agent who it is. User describes who it's working for.
Memory holds what it knows. Tools defines what it can do.
Heartbeat sets the recurring rhythm. Agents.md describes sub-agents.
Reference and project files give domain context.
Built into your frontier agent
Its identity, rules and memory settings live in the provider's app.
Human · Local Environment
User & Inputs
U
User
the human in the loop
OpenClaw Web Portal
Your Agent's App
built-in browser chat with the agent
web, desktop and phone chat with the agent
Gmail Inbox
agent reads + drafts on user's behalf
Google Drive Inbox
drop files here for the agent to ingest
Local Machine
user's laptop · code, files, Git client
GitHub Sync (from local)
push Brain edits, code, configs
Manual Brain Edits
optional — directly edit identity/memory
Public inputs (portal, Gmail, Drive) reach
You talk to the agent in the provider's
the agent over the internet. The local
app. It reaches your shared brain over a
machine connects through the encrypted
secure MCP link, so no private network
Tailscale mesh.
is needed.
The user never SSHs to a public admin
Your brain and portal still run on
port. That surface is closed.
your own server.
Tailscale
private
encrypted
mesh network
local ⇄ VPS
Linux VPS · Ubuntu Host
Provider Cloud · Their Platform
OpenClaw Runtime
Frontier Agent Runtime
Self-hosted, always-on service for the agent
Hosted and kept up to date by the provider
OpenClaw Runtime
Provider Runtime
Manages sessions, sub-agents, cron, browser, MCP connections.
Runs sessions, tools, schedules and connectors for you.
My Agent
Your Frontier Agent
the configured personal orchestrator
Claude Duet, ChatGPT Dots, Grok Bot or Muse
• Reads Brain config at startup
• Reads your shared brain over MCP
• Routes model calls in priority order
• Uses the provider's models
• Calls connected tools and MCPs
• Writes memory + outputs
• Writes notes back to your brain
• Gates risky actions through approval
• Spawns sub-agents and cron jobs
• Runs the provider's scheduled tasks
My Agent ≠ the VPS. The agent is the configured runtime living inside
Your agent ≠ your brain. The provider runs the agent;
OpenClaw on the VPS. Swap the config, and you get a different agent.
your own server runs the shared brain and portal.
OpenClaw provides the body (sessions, scheduling, browser control).
The provider supplies the body (models, sessions, schedules).
The Brain repo provides the mind (identity, memory, rules, tools).
Your brain supplies the memory (projects, people, decisions).
The two together produce a stable, persistent personal AI collaborator.
Switch providers, and your memory comes with you.
Approval & Safety Loop
Required for public, spend, irreversible, or external actions
Agent drafts
the action
Discord
In-app
approval ping
confirmation
Portal
approval page
User decides
(approve/edit/deny)
Log +
execute/abort
Connected Tools · APIs · MCPs
How the agent reaches the outside world
A · Meetings & Comms
Fathom MCP
meeting search, summaries, transcripts
Discord
notifications, conversational output
Twilio SMS
SMS backup for high-priority pings
Channels the agent uses to talk WITH the user.
B · Google Ecosystem
Google Workspace / CLI
Gmail · Calendar · Docs · Sheets · Contacts
Google Drive folders
Inbox · Processed · Reference · Projects · Scratch
User's primary work environment + the raw knowledge layer.
C · Models & Media APIs
OpenRouter API
OpenRouter Image
OpenAI API
xAI API
ElevenLabs
Cloudinary
Direct API access for image gen, speech, and hosted media.
D · Web · Data · Research
Apify
Firecrawl
Context7
Playwright
Scrape, crawl, monitor, run browser tests, look up
current SDK and API documentation.
E · Infra · Project
Firebase
Portal backend / project DB
Netlify
Portal deploys + site management
ClickUp API
task / project access
The infrastructure stack the agent operates on.
F · OpenClaw-native
Browser Control
Cron / Heartbeats
Sessions / Subagents
Nodes (paired devices)
Image / PDF / Media tools
automate the browser
scheduled + recurring jobs
child agents for background work
control paired devices
media analysis + generation
input
private Tailscale connection
sync
commit back
model call
reads identity
tool / API call
Outputs · User-Facing Surfaces
Where the agent's work shows up
Discord messages
notifications, approval pings, conversational replies
Email
to the user, plus confirmed outbound when acting on their behalf
Google Drive files
organized project outputs filed into the right folders
Agent Portal
brain, tasks, projects, people, meetings, approvals + share links
GitHub commits / pushes
to Brain, Portal, or Skills Master when approved
Cloudinary-hosted media
images, files, signed URLs for shareable assets
Twilio SMS
high-priority text backup, breaks-glass channel
Browser artifacts / screenshots
visual proof for tasks that ran headlessly
Memory · Knowledge Flow
Four layers, most curated at the top
Layer 1 · Curated memory + project status pages in the Agent Brain repo
most distilled — pinned facts, rules, decisions, and each project's "where was I" page
Layer 2 · Daily notes + assistant inbox notes
rolling log of the day, plus session notes from every assistant, folded in nightly
Layer 3 · Raw / archive — Google Drive
Inbox · Processed · Reference — ingested source material
Layer 4 · Created outputs — Google Drive
Projects · Scratch — what the agent produces, organized for retrieval
memory write
retrieval
output
memory write
retrieval
Core · Shared
Shared Brain — one memory for every assistant
Every assistant reads the same memory over one MCP link and adds short notes; the agent folds them in every night.
Every assistant
Claude app · Claude Code · Codex · Grok
· My Agent
Secure MCP door
HTTPS · each assistant signs in
separately; the owner approves it
GBrain (open source)
search · read · add a note
writes only to an inbox · no deletes
Postgres + pgvector
index of the Brain repo + inbox notes
keyword + meaning search
Every night
Screen: hold back secrets + planted text
AI curator folds notes into status pages
Commit step keeps only allowed edits
Brain repo stays the source of truth · re-indexed 5× a day
Why a shared brain
Consistent memory across platforms
→ no assistant has to be told a project twice
→ every note is signed by the assistant that wrote it
→ assistants add notes, they never overwrite
→ credentials and planted instructions held back
→ plain Markdown in git, rebuildable in minutes
→ embeddings are the only running cost
agent ⇄ shared brain (MCP)
notes fold into memory nightly
Legend · Arrow & Zone Key
input
user reaching the agent
sync
GitHub push / pull
private Tailscale
local ⇄ VPS only, encrypted
model call
agent ⇄ LLM via priority router
tool / API call
MCP or API request
output / approval
user-facing surface, gated
memory write
into Brain or Drive
retrieval
agent reading memory
shared brain
MCP reads + nightly notes
Zones
sources, tools and configuration
the agent and its memory
outputs and approvals
private network (custom agent)
grayed out: not used with a frontier agent
grayed out in Frontier mode: custom-only parts