DatumOS puts AI teammates inside the daily work of your operations and engineering teams. They pick up requests, investigate, prepare the fix and stop for a person's approval before anything irreversible happens. All of it on the AI licences your company already pays for, in your own isolated instance, with your Microsoft 365 identity.
Early access. Write to sales@datumos.ai and we set up a dedicated instance for your organisation, on AWS in the EU.
Datum · AI Agent
#platform-ops
payments-eu-2 hit its connection pool limit (64/64) at 14:01, 3 pods are restarting in a loop. Recommendation: raise the pool limit to 128 and restart the service.
#482 opened: pool limit 64 → 128payments-eu-2payments-eu-2 restarted in 41 s · 502s stopped · PR #482 waiting for review
A chatbot on a ticket queue saves minutes. An agent that owns the case, investigates with real access and hands you a reviewed fix saves the afternoon. DatumOS gives every request a structure, gives the agent the tools to work inside it, and keeps a person in charge of every consequential decision.
DatumOS works with Anthropic Claude and OpenAI models on your company's own subscriptions and API keys. No markup on model usage and no lock-in to one vendor: you choose the model per team, see the cost of every run, and decide which providers may see which team's data.
People sign in with the accounts they already have, ask the DatumOS bot in a Teams channel or chat, and hand it a shared mailbox in Outlook. The bot and connectors are installed in your tenant, so your identity and access policies apply from day one.
Each request becomes a case with a visible status, a named person accountable for it and a live summary the agent keeps current. The process rules are enforced by the platform, not by habit: no case enters investigation without a coordinator, none closes without a root cause. Managers see the whole queue, not a chat history.
The agent runs commands, reads your repositories and logs and prepares changes in an isolated environment you choose: our sandbox, a shared team machine, or a machine inside your own network. Never on the shared platform, whichever model is behind it.
Schedule recurring work, or design a Flow: several agents working in parallel, joined by checks and approval steps wherever a result carries consequences. A nightly health review or a complete feature delivery pipeline, run the same way every time.
Give each team its own space with separate data, integrations and access rules. Publish status pages your customers can read. Feed the agent your documentation and knowledge bases, so it answers with your facts, not general ones.
Mention the agent in Teams, in a GitHub pull request, in a Jira ticket or by email. It answers in the same thread, and only there, so information never lands in the wrong channel.
Everything in DatumOS can be configured through a public API, a command-line tool, a Terraform provider and an MCP server. Your engineers keep the configuration in a repository and let coding agents such as Claude Code or Codex change it, with the same review as any other infrastructure.
One platform, three places it pays off first.
Support, on-call, platform and ad operations share one reality: requests arrive all day and someone has to act on each one. DatumOS turns that stream into owned cases and puts an agent to work on them, from first triage to a fix ready for sign-off. Your people handle the decisions, not the legwork.
Your engineers stop hand-delivering every change and start designing the agents and Flows that deliver them: repeatable pipelines that carry a change from ticket to reviewed pull request, with approvals where you want them. Capacity grows without the next hiring round.
Sales leaders get answers without waiting for someone to build a report. Ask for the pipeline, a customer's history or last quarter's numbers in a Teams chat; the agent reads Salesforce with that person's permissions and answers in the same thread.
Install the Teams bot, the GitHub App or the Jira connection in your own tenant. A mention or a scheduled trigger starts the work, in the thread where it was asked.
It reads the case, the systems and the documentation you granted, works in an isolated environment and keeps a live summary as it goes. It gets only the access a task needs, and only to the systems you allow.
The answer lands where the question was asked. Anything else with consequences is either reversible or waits for approval, at the autonomy level you set per team: you see the change, the cost and the audit trail, then decide.
An agent that acts on emails, tickets and code will sooner or later be fed content designed to mislead it. DatumOS is designed on that assumption, so your security team does not have to rely on hope.
Write to sales@datumos.ai. We set up a dedicated instance for your organisation, connect the first channel together with your team and hand over the keys. During early access we onboard a small number of organisations at a time and reply within two business days.
Yes. Every organisation runs its own DatumOS instance on AWS with dedicated data storage, an encryption key of its own and its own agent runtime. Nothing is shared with other customers. EU hosting is the default; Enterprise customers can choose any AWS region.
Whichever you choose. Today DatumOS supports Anthropic Claude (Fable, Opus, Sonnet) and OpenAI GPT models, on your company's own subscriptions and API keys, so you pay your provider directly and reuse licences you already hold. The model is chosen per person or per team, every run shows its cost, and a team can restrict which providers may process its data.
Yes. Everything is available through a public API, a command-line tool, a Terraform provider and an MCP server, so your engineers can keep the whole configuration in a repository and let coding agents such as Claude Code or Codex change it through pull requests, with the same review as any other infrastructure.
Yes, that is the primary setup. People sign in with their Microsoft 365 accounts, roles are mapped from your Entra ID groups, and the Teams bot and the Outlook mailbox connector are registered in your tenant, not ours. Organisations outside Microsoft 365 can connect any OpenID Connect provider. Our team sets this up with you during onboarding.
From the panel or the API at any time: cases, sessions, configuration and attachments. A configuration export lets you move a team between instances, and case events can stream into your own data lake. Your data stays yours.
Tell us about your operations or engineering team and we prepare a dedicated instance. Connect one channel, hand the agent one real case and judge the result with the evidence in front of you.