Kanvas Update — September 2026

It’s been a little while since our last update.

In July, we introduced more structure around digital employees: roles, reporting lines, projects, and the ability to route work to the right person or agent.

Over August and September, we kept building on that foundation.

The focus has been making agents more useful in the day-to-day operation of a business. Giving them more ways to access information, carry work forward, and involve a person when a decision needs one.

There’s quite a bit to catch up on.

Work That Continues After the Conversation

A lot of business work happens on a schedule. Someone checks the numbers every morning, follows up next week, or prepares a report at the end of the month.

Kanvas agents can now be scheduled to perform tasks at a future time or on a recurring basis.

That could mean reviewing overdue invoices every Monday or preparing a daily summary from the information the agent has access to. When the scheduled time arrives, the agent runs the instruction using its available tools.

Schedules can also be paused and resumed.

We expanded support for longer-running external tasks as well. For supported integrations, Kanvas can track the job in the background and bring the agent back into the conversation when the result is ready.

These are practical pieces of infrastructure, but they matter. They let us build processes that continue beyond the moment someone sends a message.

More Tools Connected to the Nervous System

One of the biggest additions in September was support for connecting agents to external tools through MCP.

MCP is a standard that lets an agent work with tools exposed by another service. For Kanvas, it gives us another way to connect digital employees to the systems a business already uses.

We added the connection and authentication infrastructure, a catalog of MCP integrations, and controls for assigning tools to individual agents.

The catalog includes services such as GitHub, Linear, Notion, and Google Workspace. Each connection still needs the appropriate account access and configuration.

This fits the direction we’ve been taking with the Kanvas Nervous System: connect the business, define the responsibilities, and give each agent the capabilities its job requires.

As more services expose tools this way, we can expand what agents do without building every connection from scratch.

A Company Brain With Broader Context

We also introduced the Company Brain agent.

The idea is to give the team a place to ask questions that cross different parts of the business.

A sales question might need customer history and project activity. An operational question might involve inventory and financial records. The information is connected, even when the software storing it is separate.

The Company Brain brings together a curated set of reporting and lookup tools, with additional capabilities depending on the modules and permissions configured for the company.

Its role is to gather the information, identify what needs attention, and help the team decide what to do next. Specialist agents remain responsible for their own areas of work.

There is still a lot to improve here. The usefulness of that broader view depends on the quality of the underlying records and the systems we connect.

More of the Finance Process Connected

Apex and Arc, our accounts payable and accounts receivable agents, received several improvements across these two months.

They can work with a connected Gmail inbox, download invoice attachments, extract information from the documents, and use that information in their accounting workflows.

We improved support for bills with multiple line items, due dates, notes, and source attachments. Those details are easy to overlook in a demo. They become important when someone needs to review the actual bill.

We also expanded approval flows.

Pending bills and invoices can be brought into Slack for review, including the supporting PDF. Approvers can be associated with the relevant vendor or customer, and approval checks use the identity of the person making the decision.

On the receivables side, we added a workflow for extracting credit requests from attached forms and creating credit memos.

The exact process depends on how the company is configured. Our goal is to keep the document, the accounting record, and the decision connected as the work moves between people and systems.

Expenses Become Part of the Conversation

We extended the accounting tools to employee expenses.

An employee can submit an expense they paid personally, attach the receipt, and send it for approval. The tools also distinguish between money an employee is owed and an expense already paid on a company card.

That distinction matters. A record needs to reflect who actually paid and whether reimbursement is still outstanding.

This gives us another recurring administrative process that can happen through an agent while remaining connected to the accounting records.

Agents Can Work Through a Browser

APIs cover a lot, but businesses still depend on websites and portals.

In September, we added browser automation integrations and support for tracking browser tasks that run in the background. For supported providers, files produced or downloaded during that work can be saved into Kanvas and attached to the related plan.

This expands the kinds of processes we can build around agents.

A browser task still needs a clear scope, the right access, and a workflow we can verify. We now have more of the infrastructure to connect that work to the rest of the operation.

Coding Work Gets a Clearer Place in Kanvas

We continued expanding the coding agent infrastructure, including support for OpenCode alongside the existing Claude and pi.dev paths.

A coding request can be recorded as a plan and task, connected to an allowed repository, and tracked while the coding session runs.

We also added support for passing reference files into those sessions and continuing existing work.

This brings coding tasks into the same operational structure we’re building for other work: a request, an owner, a record of progress, and a result the team can review.

More Connections and Better Reporting

Alongside the agent work, we added and improved several parts of the platform:

• Odoo integration for importing and pushing contacts, organizations, and leads.
• Yusen inventory comparisons to report differences against Kanvas and, when configured, NetSuite.
• Reporting infrastructure that gives agents a defined way to query supported reporting models.
• Expanded reporting tools for events, participation, evaluations, and company activity.
• Configurable project board columns so teams can organize work around their own process.
• Better document extraction for Word, Excel, and PowerPoint files used as knowledge sources.

We also spent time on the less visible work: contact matching, communication preferences, voice-agent context, scheduled execution, and integration reliability.

Those improvements are part of making the larger features dependable enough to use.

Where We’re Going

Next week, we’re officially launching Kanvas as the harness companies can use to put agents to work inside their operations, connected to their systems, data, workflows, and teams.

We’re also preparing a specific offering for agencies that want to partner with us and bring this infrastructure to their customers. Agencies can build around the processes their customers need, with Kanvas providing the foundation for connecting systems and running agents.

We’ll share more about the launch and the agency partnership next week.