What Is Lava AI? The lava.so Desktop Agent Explained
Lava is a desktop AI agent from Lava, the San Francisco startup that Mitchell Jones founded as an AI payments company. Lava Desktop, a Mac and Windows app, puts your work apps in one window as live, signed-in tabs, with an agent beside them that reads what is on screen and acts through each app’s API. The pitch on lava.so is one line: not a chatbot, not a different AI per app, the same agent across all your software.
It is a different animal from the personal agents that launched this month. Meta’s Muse, Instinct and Wajo’s Fo run errands. Lava is aimed at the person who moves between Gmail, Slack, Salesforce and a notes tool all day and wants one sentence of intent to cover the whole loop.
Where Lava came from
Lava started as a billing layer for AI. TechCrunch reported the $5.8 million seed in August 2025, led by Lerer Hippeau with Harlem Capital, Streamlined Ventures and Westbound, for a wallet that lets agents pay for models and APIs from one balance. Jones, who studied at Yale and worked at Dropbox and Meta, had previously founded the Y Combinator fintech Lendtable.
That gateway is still the company’s core product, and it explains how the agent is priced. Lava Desktop arrived on June 26, 2026, as the consumer face of the same infrastructure. The Lava Agent in Slack followed on September 22, and there is a Lava MCP that connects Claude or ChatGPT to the same services.
What Lava Desktop actually does
Install the app, sign in, and connect your apps. Each one opens as a live tab in a sidebar. The agent panel sits on the right, and you type a task in plain language: summarize this thread, draft a reply and update Salesforce.
Two mechanisms do the work:
- Agent Connectors. To send an email, create a calendar event, post to Slack or update a CRM record, the agent calls the app’s API through Lava’s gateway. This is where most real work happens.
- The live page. The agent can read the tab in front of you for context, an open email thread or a contact record, and show the result back on that tab. Lava’s launch post calls this DOM access and says most agents have either the API side or the screen side, not both.
The connector list in the documentation is short and work-shaped: Gmail, Google Calendar, Slack, Salesforce, HubSpot, Affinity, QuickBooks, LinkedIn, and the meeting notes tools Grain, Fireflies and Granola. Lava’s launch post claims access to more than 400 API providers through the gateway, and enterprises can add internal sites the agent works across like any other app.
The agent is model-agnostic. You pick Claude, GPT-4o, Gemini or Grok and can switch mid-task. Repeated tasks can be saved as workflows and triggered from the input bar with a slash.
Approvals and modes
In the default Normal mode, the agent works through a task on its own and pauses only before high-stakes actions: sending a message or email, deleting data, making a payment, or sharing something externally. It shows an Approve action card describing what it is about to do, and nothing runs until you click. Careful mode pauses before every write. Auto mode runs the whole task with no pauses.
Lava’s own safety page is blunt about this: approving an action means you take responsibility for it, and the agent’s description of what it is about to do can be wrong.
The Slack agent works the same way. It runs on your own Lava account, not a shared bot, so installing it in a workspace gives nobody access to anyone else’s tools. Approvals render in the Slack thread, and only the person who owns the run can approve them.
Lava pricing
There is no subscription. Lava bills per call from a prepaid wallet.
| What you pay for | How it is charged |
|---|---|
| Model calls | Every task uses at least one, billed against your balance |
| Provider calls | Each API request to a connected app charges a small amount per call |
| Free providers | No provider charge, only the model call |
| Funding | Deposits from $10 to $10,000 through the dashboard |
| Enterprise | Provisioning, role-based access and shared workflows, priced on a call |
Lava’s documentation is honest that cost per task varies with complexity: a task that reads one inbox and drafts one reply costs less than one that searches three apps and makes several writes. If the balance is too low to start a task, the agent pauses and asks you to top up, and a paid call that would exceed the balance mid-task stops the task at that point. Lava has not published a typical monthly figure.
What the independent reviewers say
assistantbenchmark.com lists Lava with its funding and channels but has not run a single scored test on it yet, which is unusual for a product that launched in June. The user feedback it has collected is thin, one piece of praise for the Slack agent and a handful of use-case mentions. Treat Lava as a product with a strong architecture argument and not much public evidence of day-to-day reliability yet.
Where Lava fits
Lava is a good fit if you want to watch an agent work inside your real apps on a desktop, choose your own model, and pay by the call. The tradeoffs are the ones its design implies: the work happens in a desktop app you keep open, each send waits for you, and the monthly cost is whatever your tasks add up to.
If the job is your inbox, your calendar and the follow-up that runs whether or not you are at your desk, Carly covers it from your email, a text or the web, no app to install. It sends from your own address, starts workflows on events like a new lead or a meeting ending, and connects to thousands of apps through official connections. Free Zapier-style workflows; AI agents from $35/month. Get started here.
For the wider field, see Lava alternatives, and for the head-to-head, Carly vs Lava.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
