What Unflood does for your team
-
Agent Core turns your AI conversations into ready-to-use agents
Take the agent you built in Claude or Codex, or personalise one someone else built, and put it in production: tested, versioned and released with a person's approval. ChatGPT is next; Copilot and Gemini Enterprise come with enterprise pilots.
Explore the Agent Core -
The Library lets you discover all agents, share yours and adopt others
Give your team a Share button for agents. Colleagues adopt a shared agent in the tool they already use, with access and versions governed.
Explore The Library -
The Control System manages agent data access, token waste and AI risk
Decide who uses each agent and what data it touches; every change approved by a person and logged. Unflood maps that evidence to the EU AI Act, GDPR, ISO 42001 and the OWASP Agentic Top 10, ready for your auditor. Unflood shows the evidence; your auditor certifies.
Explore The Control System -
The Studio helps you personalise agents and run them
Start from an agent someone else built and adapt it to how you work. Your version is tested, versioned and released with a person's approval.
Explore The Studio
Discover Unflood, one idea at a time
Four short lessons. Use the arrows to move through them.
Your agents, drawn as one map
Agent Core is the map of everything an agent relies on: its agents, jobs, data, tests and releases. Unflood builds it from your chats, files and existing agents, so nothing hides inside a prompt.
- Built from what you already have
- Every dependency visible: agents, data and checks
- One core, shared with Claude and other hosts
Discover Unflood, one idea at a time
Four short lessons. Use the arrows to move through them.
Your agents, drawn as one map
Agent Core is the map of everything an agent relies on: its agents, jobs, data, tests and releases. Unflood builds it from your chats, files and existing agents, so nothing hides inside a prompt.
- Built from what you already have
- Every dependency visible: agents, data and checks
- One core, shared with Claude and other hosts
Discover Unflood, one idea at a time
Four short lessons. Use the arrows to move through them.
Prove it before anyone relies on it
The Unflood test lab checks every version before it ships. Evaluations measure the answers, guardrails set the limits, a panel of judges reviews the result and a gate stops a release that fails.
- Failures become items to fix
- Each release carries its evidence
- A person approves; Unflood never decides alone
- 1EvaluationsCases that score the answers, run more than once.
- 2GuardrailsLimits on cost, data and actions while it runs.
- 3Panel of judgesSeveral judges review the result and give one verdict.
- 4GatesA release needs a pass and a person to approve it.
Sample lesson: what each part does, in the order a version meets it.
Discover Unflood, one idea at a time
Four short lessons. Use the arrows to move through them.
Learn what governance asks of your agents
AI and agent governance is hard to read. Unflood explains what each framework asks in plain words, and attaches the test evidence that goes with it. It is evidence for your own review, not a certification.
- Plain words, no legal jargon
- Evidence from your own tests and releases
- Starts with the frameworks most teams meet first
- 1EU AI ActRisk classes and the duties that come with them.
- 2OWASP agentic risksThe known ways agents go wrong, and how to test for them.
- 3ISO and NISTThe management frameworks auditors ask about.
- 4GDPRWhat an agent may see and keep about people.
Unflood provides evidence for your review. It does not certify compliance. Lawyer review of the final wording is pending.
Discover Unflood, one idea at a time
Four short lessons. Use the arrows to move through them.
Make an agent yours, faster
Start from an agent someone else built and adapt it to how you work. Unflood boosts the adaptation: it proposes changes you accept or refuse, tests your version and releases it only with your approval.
- Your version stays separate from the original
- Proposed changes are marked as proposed
- Tested and approved like any other release
- 1Start from a shared agentPick one in the Library that is close to what you need.
- 2Adapt itChange it in chat; Unflood proposes edits you accept or refuse.
- 3Test your versionThe same checks run on your changes.
- 4Release with approvalA person approves before your version is shared.
Sample lesson. Personalising is being built; this shows the intended steps.
How it works
One agent, from the chat where you built it to the colleagues who use it.
-
Step 1
Build in Claude
Describe the agent in a Claude chat. Unflood records it, its tools and its tests.
-
Step 2
Test it
Suites and a judge run it; each failed case becomes an item to fix.
-
Step 3
Release it
Package a version. Nothing reaches people without a person's approval.
-
Step 4
Share it
Choose who can use it, person by person or team by team.
-
Step 5
Evolve it
Change it, retest it, release the next version. Every version can be rolled back.
Works with the tools you already use
One connection serves every host. Unflood decides which agent may use what, and keeps the evidence.
Where you build
Start in the chat or tool you already use.
- Claude Anthropic
- Codex OpenAI
- ChatGPT OpenAI
Where you run
Agents run in your own account, never on Unflood.
- Claude Anthropic
- Microsoft 365 Copilot Microsoft
- Gemini Enterprise Google
- Azure AI Foundry Microsoft
Tools you work with
Through each vendor's own connector.
- Microsoft 365 SharePoint, OneDrive, Outlook, Teams
- Google Workspace Gmail, Drive, Calendar
- Jira and Confluence
- GitHub
- Salesforce
- ServiceNow
- Slack
- Dataverse
- Snowflake
- Box
Today Claude works end to end. Codex is built and not yet checked live; ChatGPT, Copilot, Gemini and Azure follow. For tools, file storage for SharePoint, OneDrive and Google Drive is built and tested on recorded answers; agent tools are planned.
Why Unflood
What a chat alone can't give you.
-
Governed releases
Package, publish and promote with gates. Nothing reaches users without a person's go.
-
Tests as proof
Suites, graders, a judge and repeated runs, with failures raised as items to fix.
-
Portable, no lock-in
Unflood never runs your agents. They stay where your people already work.
-
Audit and evidence
Lineage, evidence, impact, cost and coverage for every agent.
You stay in control
- A person approves every release
- Every version can be rolled back
- Access per person, with delegation
Questions
Is Unflood a workflow or automation tool?
No. Workflow tools run steps you define. Unflood is a crew of AI agents that understands your agent, writes its tests, reviews its design and explains failures in the background, and leaves every decision to a person. The crew is being built; each card above shows its status.
How is Unflood different from eval or observability tools?
Those measure an agent. Unflood takes it from the chat where you built it to a tested release your colleagues can use, and works alongside them.
Does Unflood run my agents?
No. Your agents run where your people already work. Unflood holds their definitions, tests, evidence and releases.
Which hosts work today?
Claude, in a private pilot. ChatGPT is next; Microsoft 365 Copilot and Gemini Enterprise come on an enterprise pilot.
Who approves a release?
Always a person. Unflood's own agents propose; they never approve.
Bring your first agent to your team.
We onboard each pilot team ourselves.