The Pulse
Stitch brings AI agents into its banking operating system
Stitch AI offers bank staff a copilot, ready-made agents and a framework for building their own agents. The company says the tools operate within bank-set permissions and approval controls.

Banks using Stitch’s operating system can now use AI agents to retrieve data, carry out tasks and change configurations under bank-set controls. Stitch introduced the product on Nov. 17 and described its features in a release the following day.
Source prnewswire.com
The release describes three ways to use Stitch AI: a copilot for staff, ready-made agents for banking operations and a framework for building custom agents. The company says the tools run inside its operating system rather than through separate connectors.
Copilot and ready-made agents
The copilot is designed to take plain-language requests from bank staff. It can pull account and customer data, set up products ranging from accounts and cards to loans, and update workflows, according to Stitch. Product configuration can include limits, pricing, fees and eligibility.
Every change made through the copilot waits for approval, the product page says. Staff can also use it to ask for data or request a workflow or pricing change.
The ready-made Stitch Agents are intended for work including onboarding, credit decisioning and collections. They operate within a bank’s existing workflows, permissions and rules, the company says. The product page also lists agents for fraud checks and customer service.
Building and controlling agents
The Agent Harness lets banks build, configure, test and deploy their own agents. It includes permissions, approvals, guardrails and audit trails, according to the release. On its product page, Stitch says teams can choose which services and systems an agent may use before testing and publishing it.
All three parts of the offering operate within limits set by the institution and ask when an action exceeds those limits, Stitch says. Banks decide what an agent can do independently and what requires sign-off.
The company says each agent has its own identity, scoped permissions, hard limits, a maker-checker control and an audit trail. Sensitive information, including identification numbers, account numbers, phone numbers and addresses, is masked before a model sees it. Questions, actions and approvals are logged along with their reasoning, according to the release.
Deployment choices
Institutions choose where Stitch AI runs and which model it uses. The release says it can be deployed in-country or in a private cloud, and that banks can switch models without rebuilding the setup. The product page lists open, closed and custom models as options.
Stitch says the agents act on the core system through the same services used by the bank’s products, without copies of data or separate connectors. The company contrasts that approach with AI tools added on top of older banking systems, which it says can work with incomplete copies of records and lack direct access to accounts or customers.
Company background
Stitch’s platform spans financial products including lending, cards, payments and wallets. The company was founded in 2022 and is headquartered in Riyadh, according to the release. It says it serves financial institutions around the world and is backed by investors including Andreessen Horowitz.
Sources
Related stories
- Google opens global waitlist for experimental Stitch Loop agent
- Most banks in BCG benchmark permit AI agents, but few have full controls
- Google brings Hydrogen robotics tools to paid Antigravity users
- NVIDIA expands CUDA-X tools for scientific AI agents
- Reflection restricts Laser penetration-testing preview to vetted users