An open-source AI agent & workflow runtime. Build with any AI framework, it runs with Soma.
Security, SaaS integrations, approval workflows and debugging tools for better DX.
✔ Created typescript agent from example template.Features
An OSS runtime
All of Soma's features are open-source and self-hostable.
A data & governance plane
A single view across all your AI agents messages to users, tools and model requests.
Secure by default
Encrypt credentials and proxy them in to tool calls to avoid exposure. Protect agent endpoints with Oauth, API keys or OIDC.
Custom triggers
Deploy to where your users are in Gmail, MS Teams, Slack and more.
Tool Integration
Integrate 100's of SaaS MCP tools from our marketplace or build your own.
Secret & virtual MFA vault
Share secrets between tools without leaking them to model providers.
Automatic observability
Logs and traces stored and browsable. An LLM gateway to monitor all outbound model requests.
Embedded datalake
Sync SaaS data into a SQL datalake. Surface insights with ease.
Agent orchestration
Build & deploy a network of agents. Route tasks to the best agent for the job from a single inbox.
Our vision
Agents are the new UX for workflow integrations & automation
The time for big clunky ERP software or brittle SaaS integrations is over.
For a long time, building highly dynamic workflows to automate business processes has been unachievable for small to medium businesses. Building internal tools that are dyanmic and personalised has required teams of internal engineers or hudreds of thousands of dollars to ERP implementation consultants.
Agentic AI provides a 10x user experience improvement to what's available at a cost that any scale up can take on.
Don't sweat the small stuff. Let us take care of the infrastructure and tool integrations so that you can focus on automating workflows based on your needs and surfacing them effectively.
A future with many specialized agents, orchestrated together
The term "agent" is loaded. Put simply, we see agents as dynamic code, able to adapt to inputs that traditional if statements and loops never will.
The more specialized they are (i.e. an agent that processes expenses from emails and uploads them to your accounting software), the more effective they are than a single generalist agent (i.e. "an AI sales co-pilot")
Because of this fact and future downward pressure on foundational model inference costs, we expect to see the average small to medium business needing to manage deployments of 10+ agents. We're providing specialized observability, monitoring and tooling to help monitor and manage these deployments.
If you've made it this far and you're interested in shaping the future of software engineering, apply to join our team as a co-founder.
Join founding team