Always-on agents that get smarter over time. Open source, MIT.
SERVICES AGENT PRODUCTS
You already have a product. Your users want something inside it that remembers them, calls your tools, and keeps working when nobody is watching the tab. We build that on Stratus Agent, in your codebase, and hand you the keys.
It is still there tomorrow.
Sessions that survive a restart, memory the agent writes and reads on purpose rather than a transcript it drags along, and schedules so it can act before it is spoken to. Closing the tab does not reset your customer’s relationship with it.
Every capability declared, nothing else reachable.
Your endpoints become tools the agent may call. Each agent declares its own list and nothing outside it exists. Writes, shell and network stop for approval until you widen the policy yourself, per tool and per path, and an approval that should outlive the session is recorded with who granted it.
Slack, a dashboard, an API, MCP.
A per-agent Slack identity with its own avatar and presence, a control API with a web dashboard on top of it, and an MCP bridge that mounts any existing MCP server as tools under the same policy. You pick the ones your product needs and skip the rest.
Tokens attributed per agent, provider and model.
Usage accounting is part of the runtime rather than a log you parse afterwards, so per-plan limits and per-customer invoices start from real records. Prompt caching is on the stable prefix, so the same context does not cost full price every turn.
The gateway, the permission engine, Slack, the control API and dashboard, tools for files, shell, web and browser, MCP, memory, schedules, and usage accounting. Running today, in the open.
Hosted multi-tenancy, per-agent process isolation, and the fleet console are on the roadmap and not in your install yet. We will tell you which side of that line your product needs before you sign anything.
Stratus Agent is MIT licensed and the roadmap is public, one written spec per step with its real status. You do not have to take our word for either list.
View on GitHub ↗01
What the agent does for your users, which of your systems it touches, and what it must never do. The third one takes the longest and is the one that matters.
02
We work in your codebase and through your review process. The agent, the tools, the policy, and the surface your customers actually see.
03
It runs on your infrastructure under your provider account, and every line of it is in your repository. We stay on afterwards if you want us to.
If the answer is something your product already does, this is a short project.