CONVERSATION MEMORY
What should carry forward?
Past messages, facts and decisions give an agent continuity.
The way an explanation clicks for you deserves a place in your AI context.
Articulate turns available local coding activity into working preferences you review—so your next agent has a starting point for how to explain, build and collaborate with you.
Free testing release. Built for feedback, still evolving.
Codex Claude Code Claude Desktop Cursor
CONNECTED THROUGH MCPTHE IDEA BEHIND ARTICULATE
Imagine ten people reading the same technical guide. One asks for an example. Another wants the steps. A third wants the reasoning behind the decision.
Think about the last time something finally made sense. What helped: a concrete example, a connection to work you already knew, or a chance to try it yourself?
That is the motivation behind Articulate: give you and your AI a clearer, shared starting point for how to work through things together. Your preferences can change with the task, the project and what you already know.
Today, the local pilot starts with requests you have actually made in supported coding-agent history. You review what it finds and decide what travels to another agent.
CONVERSATION MEMORY
Past messages, facts and decisions give an agent continuity.
ARTICULATE’S FOCUS
Reviewed preferences add guidance for explaining and collaborating—such as walking through an example before the abstraction, or showing the cause before a fix.
Memory tools can support personalization too. Our focus is making your reviewed working preferences portable across agents. The pilot does not measure intelligence or claim proven learning gains.
THE FLOW
Start with one project or your work across projects. Choose your agents when you’re ready to connect.
Find supported local activity from Codex, Claude Code and Cursor. Articulate can use the traces available on your device; cloud-only and deleted chats may be missing.
Check suggested preferences, edit them, or keep your own. The default scan finds explicitly stated requests. Broader interpretation needs a configured local model.
Choose which agents can receive your reviewed preferences. Each connection needs its own approval. Use profiles and saved workflows for different kinds of work.
LOCAL BY DESIGN
A profile should be useful to you.
And understandable to you.
Raw imported activity is not returned through Articulate’s MCP tools. Connected AI providers can process the reviewed context you choose to share.
They guide how an agent presents its work. Your current task, facts and safety still take priority. A stored preference is not proof of a better answer.
Review profiles, revoke an agent’s access, or erase Articulate’s local data through the CLI. Removing a package and deleting saved data are separate steps.
SMALL START. REAL FEEDBACK.
Install the public pilot, run the guided setup, and try one real task with an agent you use.
node --version and npm --version. Check your version and get help ↗Local pilot for macOS, Windows and Linux. Keep an existing project’s Node setting; use a supported runtime in the terminal running Articulate.
npm i -g articulate-local@pilotarticulate One project
My work across projects
Connect my agents
Then run articulate in your project.
↑ / ↓ choose · Enter confirm · You review before sharing
The public pilot is available on npm. Browser account connection is being tested separately; required CLI/plugin sign-in is not in the published pilot. No paid plan or Pro trial is running.
HELP SHAPE WHAT’S NEXT
Describe the experience. Keep secrets and private conversations out of your report.