KummaLabs builds, connects, and fixes AI agents for email, leads, documents, phone, and internal tools. Each one reads the request, uses your data, takes the action, checks that it worked, and hands off to a person when it should.
We work in English and 中文.
Customer email
“Can you quote 40 branded mugs for an event on the 18th? Same logo as last time.”
Read the request (Done)
40 mugs, branded, needed by the 18th, repeat customer.
Found the customer in the CRM (Done)
Previous order and logo file on record.
Priced it from your price list (Done)
Quantity pricing came from the product table, not the model.
A detail was missing (Needed attention)
No delivery address on file, so the reply asks for one.
Reply drafted for approval (Waiting on a person)
A person reviews it and clicks send.
Result in the CRM
An agent is only useful if the result shows up in a system you already use. The last column links to something we’ve built that does the same kind of work.
Reads: Customer emails and attachments
Does: Sorts them, looks up the customer, drafts replies for a person to approve
Writes to: Email, CRM, help desk
Not yet
Reads: Form fills, inquiries, lead lists
Does: Researches, qualifies, drafts follow-ups, routes to the right person
Writes to: CRM, calendar, Slack
Reads: PDFs, photos, spreadsheets
Does: Extracts the fields, checks them against your records, flags what doesn’t match
Writes to: Database, spreadsheet, review queue
Reads: Calls, SMS, WhatsApp
Does: Answers, takes orders or requests, confirms by text, transfers to staff
Writes to: Order system, database, SMS
Reads: Your APIs, databases, and MCP servers
Does: Runs tasks through scoped tools and asks before risky actions
Writes to: Your internal systems
Reads: Websites, search results, feeds
Does: Collects and grades sources, and rejects citations it didn’t actually read
Writes to: Reports, database, alerts
A new AI agent for one business process.
Typical first project: One process, end to end, with a person approving anything that matters.
Connect an agent you already have to the tools you already use.
Typical first project: One integration, with duplicate and failure handling, tested on your edge cases.
Find out why an agent or automation misbehaves, and fix it.
Typical first project: A fixed-scope audit: we run scripted failure cases and give you a written fix list.
Prototypes, internal tools, and open source we’ve built. Each one is labeled for what it is. None is presented as a client deployment.
Phone ordering agent, English and Mandarin
Voice, typed tool calls against a database, confirmations that are checked, and a hand-off to staff.
Prototype
Document extraction with validation
Turning photos and documents into structured data, with checks that remove anything the source doesn’t support.
Prototype
Research and monitoring
An agent whose citations can be trusted: it only counts sources it actually read.
Internal tool
Lead research and follow-up drafts
AI drafting with rule checks and a person in the loop before anything leaves.
Internal tool
Agent runtime and orchestration
The plumbing behind agents: tool registries, worker delegation, MCP, memory, and approvals.
Open source
What happens today, which systems are involved, and what a good result looks like. A few lines by email is enough to start.
Deliverables, exclusions, acceptance tests, milestones, and price, agreed in writing before any work starts.
You see working pieces in your own systems and pay milestone by milestone.
Documentation, tests, and everything running in your accounts. Support terms are agreed up front, not open-ended.
Email a few lines about the process you want to automate, connect, or fix. You’ll get back questions or a proposed scope. Agencies that need delivery help are welcome too.
Email [email protected]