Why Starlings
Starlings is where your people and your agents work on the same record: every call, message and task, filed on the client it’s about. Meetings write themselves down without being recorded. Your agents get a seat, not a login: they work on the same records as your people, under the name of the person who asked, and Starlings logs everything they touch.
Built for companies and crews of 2 to 50. Host from a Mac or iPhone.
Simulated views of Starlings with sample data.
Who it is for
Consultancies, studios, research and advisory teams, small funds, and new companies and crews starting something of their own: 2 to 50 people whose work is conversations with clients, partners and each other, and who already hand part of it to Claude or another agent. Whoever runs the call hosts from a Mac or iPhone. Everyone else joins from a browser. The team signs in with Google Workspace.
Sound familiar?
The client call is on Zoom. Fireflies posts the transcript to Slack. A Zap opens the Linear issue. The deck is in Drive, the contact is in HubSpot, and an MCP server on each side lets Claude read all of it. The plumbing works. What it cannot give you is a project: each tool has its own idea of what a project is, who a person is and what counts as done, and you maintain the connections between them. Two weeks later the decision is in a thread nobody can find, and the notetaker recorded the client without anyone asking.
Envisioning ran like that. Starlings is what we built instead.
One record model
Instead of one project spread across seven tools.
In Starlings a project is one record, and the conversation, the tasks, the issues, the documents and every meeting about it attach to it. A person is one identity across all of them. Anyone who opens the project sees all of it without searching seven tools. What your integrations connect today, Starlings stores in one data model.
Agents
Instead of an MCP server per tool and an agent that reassembles the company from all of them.
An agent preparing a client meeting needs the last conversation, the open tasks, the documents and the decisions already made. Starlings keeps them on the same project, so the agent can read the context and return work to the place the team will find it. You can review the result beside the records it came from.
There are two ways in. Your Claude connects to Starlings as you through Model Context Protocol (MCP), with reads and any writes you allow recorded in Starlings. A shared agent appears in Team: message it or mention it in a project’s conversation, and its reply lands there. Envisioning’s agents run on Vercel Eve, with their own tools, skills and connections. Starlings holds the shared work; the runtime does the job.
Prepare a meeting brief from the project’s history. Review what is still open. Draft a document and file it on the client. These are useful jobs because people and agents can work from the same record, and the next person can pick up where the work left off.
Direction. An agent, yours or a customer's notetaker, joins the room as a participant the owner admits. Starlings shows it, hears it and transcribes it like any other participant.
The agent ecosystem
Our direction is to let all types of agents work alongside your people: personal assistants, shared coworkers, coding and research agents, browser and computer agents, scheduled workflows, and agents you build or run yourself. Bring the agents you already use, across providers, and give them a place in the team’s shared work.
Claude, Grok Bot, Dots and Muse bring assistants. Eve, OpenClaw and Hermes provide ways to run them. They can research, use tools, operate computers and follow reusable skills. A team still needs somewhere to keep the assignment, its context and the result.
That is Starlings’s role: the client or project the work belongs to, the conversation around it, and the person on whose behalf it was done. Claude is a working connector today; Eve runs our shared agents. We are preparing common integration contracts so other providers and runtimes can participate too. The products below are examples of that ecosystem, not a closed list of agents Starlings intends to support. Their integration status separates what exists from what needs validation.
Researched on 3 October 2026 from the linked official sources. These products are examples of an open ecosystem, not a complete catalogue or a limit on which agents should work with Starlings. They supply different parts of an agent’s capabilities. This is a capability comparison, not a reliability benchmark. Eve and Claude work with Starlings today; the other rows are integration opportunities, not supported connections or delivery commitments.
| Product and role | Capabilities | Relationship to Starlings |
|---|---|---|
| Vercel EveCustom agent runtime | Durable sessions, sandboxed code, typed tools, Markdown skills, connections, schedules, subagents and traces. Browser work can be added through tools. | Working todayRuns Envisioning’s agents outside Starlings. Signed handoffs carry the assignment; replies return to the conversation. |
| ClaudeAssistant and connector client | Chat, Code and Cowork; connectors, skills, files, scheduled tasks and browser or computer use. Capabilities depend on the client and plan. | Working todayClaude connects to Starlings over MCP as you. Read company records, prepare work and make writes you have allowed. |
| Grok BotPersonal and team agents | Persistent cloud computer, browser and terminal; connectors, reusable skills, routines, memory and coordination between bots. | Integration to validateA candidate MCP client for a member’s assistant. A bot participating in Team would need a separate signed adapter. |
| OpenClawAgent runtime you operate | Messaging channels, shell, files, browser, search, skills and plugins; background automation and coordination across agents. | Integration to validateA candidate runtime for a Team agent, with an adapter for Starlings identity, handoffs and replies. |
| HermesAgent runtime with procedural learning | Terminal, files, browser automation, MCP, memory, session recall, agent-created skills, scheduled work and delegation. | Integration to validateA candidate MCP client or Team-agent runtime. Pairing and signed message delivery need separate validation. |
| OpenAI DotsAlways-on personal assistant | Cloud computer and browser, connected apps, ongoing tasks and memory; background agents and delegation to Work or Codex tasks. | Integration to validateA candidate personal connector for reading context and returning work as the member who asked. |
| Meta MusePersonal and small-business assistant | Computer, browser, terminal, files and connectors; proactive work, parallel tasks and documents, PDFs, websites and interactive outputs. | Integration to validateA candidate custom connector. We still need to test how it signs in to Starlings and how its returned work gets reviewed. |
| InstinctPersonal assistant by text and calls | Phone and computer operation, connected applications and devices, personal context and proactive follow-up. Public developer details are limited. | Integration to researchAn integration to research. A public Starlings-compatible connector or developer contract has not been established. |
| ManusAgent for delegated tasks | Cloud sandbox and browser work, files, research and generated deliverables; connectors including MCP, plus an API for external workflows. | Integration to validateA candidate connector or task adapter for assignments and completion notifications. |
| Gemini EnterpriseEnterprise agent platform | Enterprise knowledge, research and multi-step workflows; custom ADK and A2A agents, with centralized management. | Integration to validateA candidate enterprise integration. Mapping company access and the agent’s acting identity comes first. |
| Microsoft Copilot StudioEnterprise agent builder | Knowledge-grounded conversations, workflows, connectors, REST and MCP tools; agent handoffs and file work depending on the harness. | Integration to validateA candidate MCP or workflow integration for teams using Microsoft 365. |
Direction: agents working together. An Eve agent could research a client brief and hand its sources to an agent on another provider to draft it, with the result returned to the same project for a person to review. Each handoff needs an accountable person, bounded access and a clear stopping point. Cross-provider delegation is anticipated; it is not available in Starlings today.
No recording
Instead of a recording nobody watches and a bot nobody agreed to.
The hosting device transcribes locally while the call happens, with each speaker on their own audio track so names are right. Only finished text leaves it. Guests see live captions, so they know exactly what is being kept. You can tell a client "we don't record" and mean it. The same Mac transcribes your Zoom, Teams and Google Starlings calls and files them beside the rest.
Guests
Instead of an install, an account and a mic nobody enabled.
A guest opens the link, types a name and waits until you let them in. Your booking page gives them a time and the same link. Guests install nothing. Starlings issues no media credential until you admit them.
After the call
Instead of a summary bot posting into a channel.
When the call ends, Starlings files the summary on the project and on the contact. Starlings offers each commitment in the transcript as a task. An accepted task links to the meeting it came from. You can send the note to the person in one step.
Private by default
Instead of a company-wide feed.
Home shows the next meeting, the tasks due and the thread that moved. Quick Notes and your personal calendars stay private until you file them. A filed item takes the access rules of the place you file it into.
Your instance
Instead of renting your meetings from a platform that keeps every company's in one place.
The room runs from the Mac or iPhone in front of you, under your Google account and your domain. Media is relayed, never stored. Records, documents and credentials each live in a service of their own with its own store and its own key. Native apps on Mac and iOS; the browser for everyone else.
Under consideration. Today Starlings runs Envisioning. We are setting it up for a few firms we already work with, each in an instance of its own that we run and upgrade for them.