Implemented in the project
- Agent chat plus sandboxed personal apps (“Gadgets”)
- Per-workspace Durable Objects and Dynamic Worker facets
- Gatekeepers with scoped access, action logs, and human review
CinaSeek / early-access AI productivity environment.
Build and run private AI-assisted apps inside capability-based sandboxes.
CinaSeek combines agent chat, personal applications called Gadgets, and scoped Gatekeepers. The project is early access; production readiness depends on the release and deployment configuration.
Product evidence / current scope.
What is verifiable today.
Status and capabilities below are tied to public project repositories. They do not imply a customer deployment, managed SLA, or production availability beyond the linked evidence.
Status: Early access
The repository documents an August 2026 early-access build: workerd is for local development, with Cloudflare as the documented production path.
Verified: · commit fefb5a3
Documented interfaces and integrations
- Cloudflare Workers / workerd
- Cap’n Web RPC
- Gatekeepers for OAuth, MCP, GitHub, and Google when configured
CinaSeek
Example data generated by this site—not a customer screenshot or proof of a live deployment.
Input Planning request
Control Sandbox + scoped Gatekeeper
Output Private Gadget + review queue
Evidence sources
Capabilities.
A focused layer for the work your team needs to finish.
These are reference workflow patterns, not a promise that every capability is enabled in every release or deployment. Start with the smallest useful workflow and confirm current availability against the evidence above.
Work through an agent chat UI.
Ask an agent to complete tasks or build a focused personal application for the work in front of you.
Create private Gadgets.
Generate small applications whose client and server run in isolated, per-user boundaries.
Collaborate through explicit sharing.
Keep each Gadget private by default, then grant access to named users or through a deliberate share link.
Introduce only approved resources.
Use Gatekeepers to narrow authorization to the specific external resources a user intended.
Inspect actions before release.
Log external actions and preserve human approval for operations with side effects.
Develop locally, deploy through the documented path.
Use workerd for local development and the documented Cloudflare path for production deployment.
Workflow.
Move from a request to a sandboxed personal app.
A reference evaluation keeps the requested job, generated Gadget, introduced resources, and review queue visible.
01 / Define the job Describe the result, intended users, data sensitivity, and external resources required.
02 / Build the Gadget Let the agent generate the focused client/server application inside its isolated workspace.
03 / Introduce resources Connect only the approved external resources through Gatekeepers and explicit capabilities.
04 / Review and share Inspect queued side effects, test the result, and choose whether to keep it private or share it.
Illustrative Gadget contract.
Keep the implementation understandable.
Use explicit inputs, review points, and outputs so the system is easier to operate and change.
gadget "planning-workspace" {
sandbox = "isolated"
access = ["introduced-resources"]
actions = "review-required"
}Operating principles.
Built for practical adoption.
Capability boundaries.
Agents and Gadgets begin without ambient access to external resources.
Private by default.
Each Gadget runs in an isolated boundary and is shared only through an explicit user action.
Visible review.
Actions with side effects should remain logged and available for human approval.
Next step.
Map this product to one real workflow.
Tell us the work, systems, constraints, and review points your team already has. We will help identify a sensible starting scope.