Ambitious missions need a crew.
Bring frontier agents aboard.
The collaborative AI agent workspace where your team and your agents work together. Any model, any cloud, your infrastructure.
Turn recurring work into an operating policy. Then let agents run it.
Governed by policy. Define your rules, set permissions, add your tools, build your memory, and manage approvals, all in one place.
Isolated by default. Agent work runs in disposable sandboxes with only the files, tools, and network access you allow.
On the record. Every action stays on the record. Always know exactly what your agents did, and why they did it.
Connect the tools you use every day.
Build the harness around every agent. Written by you and enforced at runtime.
Collaborate anywhere. Work with colleagues and agents side by side, in the d5s workspace or straight from your apps like Slack and Teams.
@d5s the beta launch moved to Thursday, line everything up.
add the pricing update to the release notes too.
Launch prep is done.
- ✅ Release notes drafted in Notion, pricing update included
- ✅ Status page update scheduled
- ✅ 14 customers queued for outreach
Announcement goes out Thursday 9:00.
Plugs into your tools. Bring in the tools your team already uses. Permission-aware, so it only ever sees what you allow.
+280
A secure workspace for every run. Agents get a real workspace to do the work. Your secrets and systems stay sealed off. Nothing leaks across conversations or teams.
Not locked to any one model. Choose from any organization-approved LLM provider, bring your own keys, or run your self-hosted model for a fully air-gapped deployment.
Don't see your provider? Bring your own keys, or plug in the model you host yourself.
Efficiency that compounds. Agents learn how you work. Once an agent does a job well, it saves the steps as a playbook, so the same work runs faster and cheaper next time.
Working in the background. Background agents wake on a heartbeat, run their checks, and reach out when something needs you. Every clean run sharpens their playbooks.
Answers grounded in real sources. Web search gives fast, cited answers. Deep research runs a multi-step investigation across the open web and your own data, and hands back a brief with every claim linked to its source.
Ask anything
Automations on a schedule, agents on a shift. Real work, executed end to end.
Security-first by design. Built for regulated workloads.
Deploy in single tenancy, inside your own VPC, or fully on-premise, in the region you choose. It can run completely air-gapped, with no third parties.
FAQ.
d5s is the AI agent workspace for teams that want frontier AI capability without the lock-in. Any model, any cloud, your infrastructure. Set up agents that handle repetitive work for the whole team, like clearing the support queue each morning, tracking suppliers, or pulling together the weekly report. Each one plugs into your existing tools, follows rules you write in plain language, and logs every step.
Teams that want AI doing real work, not just answering questions. Best fit: legal, finance, life sciences, infrastructure, and government suppliers, where data sensitivity rules out hosted chatbots. And teams that refuse to bet the business on one model provider: they match each task to the best model for the job, or the most cost-efficient one, and switch as the frontier moves, without re-platforming.
Get in touch. We onboard new teams with a short call, help define your first workflow, and connect it to one of your systems in the first session.
Hosted chatbots stop at the reply. With d5s, agents run in isolated sandboxes with scoped connectors to your real systems: reads logged, writes gated, every action under a policy your compliance team signed off on.
Depends on who hosts it. When we host it: in AWS, in the region you pick, isolated per customer, with infrastructure, identity, and model access handled for you. When you host it: inside your own cloud, in any region, behind your own login system. Data never leaves the environment you deploy into.
No. Your data, prompts, and documents are never used to train models, ours or anyone else's. Agents read only what you connect, for the run they are doing, and nothing is kept for training.
You pick. We handle access to OpenAI, Anthropic, Mistral, and others, or you bring your own keys, including open-weight and self-hosted models. Each agent is locked to a specific model, so switching providers is always explicit and reviewable.
One package covers the control plane, sandboxes, identity, and connectors. Deploy it inside your own cloud or on-premise, next to the systems your agents need to reach. Nothing has to leave your network.
Yes. Self-hosted setups plug into the identity provider your team already uses for single sign-on and access control, including fully air-gapped deployments where no traffic leaves your network.
Only the things you connect, with the permissions you set. Databases, SharePoint folders, Slack channels, internal tools. Each connection is read-only, write, or off, and you can change or revoke it any time.
Every run is sandboxed and destroyed when it finishes. Connector access is read-only, write, or off, and every read, write, and decision is logged for audit.
Changelog. New features, improvements, and fixes shipped to d5s.
Usage analytics — by agent, by model, by skill
A new Agent stats tab in Settings breaks down runs, tokens, and spend per agent. A calendar range picker replaces the date inputs, usage charts cover models and skills, and you can download your billing data as a CSV.
Agent management — scheduled wakeups, presence, and richer chat surfaces
See scheduled wakeups at a glance on the agent rail, manage channels and Slack bindings through a redesigned dialog, and see who's online with a workspace presence facepile. External chat messages now render as proper markdown.
Deep research, redesigned — interactive sources, filtering, and a dedicated skill
Deep research results are now interactive hypermedia panels with built-in filtering and stateful navigation. A new research-report platform skill produces publication-ready reports, and stopped research sessions recover their progress.
Tell us the work
you'd hand off
We're working with our first teams now. Tell us the job you'd want an agent to do, and we'll show you what it would look like.