Gumloop

You create an agent, add the two or three connectors it needs and pick one of 35+ models, or bring your own API keys. You give it a real task and watch it run; once it works the way you want, you ask it to write its own instructions and save the approach as a skill. Agents can then run on app triggers or recurring schedules, be shared with teammates as owners, editors or users, and be called through the API, CLI or MCP. Usage is billed in credits.

• •400+ connectors - Agents read and write across apps such as Slack, Gmail, Salesforce, HubSpot and BigQuery, with app triggers and recurring tasks to start work automatically.
• •Agent browser - Agents open a real browser to click, type, upload and pull data from sites without an API, using saved logins, CAPTCHA handling and recorded sessions.
• •Self-improving agents - Each agent reviews its own runs, spots steps that were slow or cost too much, and updates its instructions and skills for next time.
• •Team sharing - One person builds an agent and the rest of the company uses it in their own chats, with role-based permissions, shared credentials and a shared credit pool.
• •Spend and governance controls - AI spend insights, evaluations against rules you set and connector guardrails, with RBAC, SCIM/SAML and audit logs on the Enterprise plan.

Revenue, marketing, support and operations teams use it to automate research, CRM updates, outreach drafts and reporting. Data and RevOps people use it to answer warehouse questions from Slack. Larger companies use the Enterprise plan for governance and security controls. With no free plan and usage billed per credit, it suits teams with a clear process to automate more than casual experimenting.