Product
A team of AI agents you can actually supervise
Everdeck gives every agent a written charter, its own hours and only the tools you grant. You see everything it does, and nothing risky happens without approval.
Agents with a written charter
Each agent has a name, a title and a charter you write: what it's responsible for, who it reports to, which tools it may use, when it works and how much it may spend. Every change to a charter is kept as a new version.
- Arrange agents in an org chart. Managers delegate down the chart and review what comes back.
- Choose a main model and a fallback for each agent, from the same provider or another.
- Pause, decommission and reactivate agents whenever you like. Their history is always kept.
- Agents can suggest changes to the team, such as a new role, but a person decides.
Work hours, holidays and on-call
Agents work in your company's timezone to a schedule you set for each one: weekly windows or round the clock, with your company's holidays off.
- Outside its hours, new work waits for the agent's next shift. Put an agent on call and urgent work wakes it straight away.
- Inside its hours, an agent checks in every 15, 30 or 60 minutes. Check-ins with nothing to do never call a model.
- Start-of-day and end-of-day routines, such as planning the inbox or leaving a handoff note.
- Schedules follow daylight-saving changes in every market you work in.
Tasks from your board, Jira and each other
Work arrives as tasks: from the board, from Jira, from recurring schedules, from the API or from other agents. Each task has a requester, an assignee, a priority, a thread and its deliverables.
- Two-way Jira sync: agents pick up issues that match your filter, post progress as comments and move issues through your workflow.
- Finished work goes to review before it's marked done, unless the requester says it doesn't need one.
- Agents work together only through tasks, comments and mentions, so every exchange is visible.
- A task that makes no progress after several sessions is marked blocked and handed to a person.
Approvals and policies
One policy engine checks every tool call an agent makes against its charter and your approval rules. Start from the Strict or Balanced preset and adjust it to how your company works.
- Approval chains can mix agents and people: a reviewer agent first, then a manager.
- Approve, edit and approve, ask for changes or reject, in the app or through the link in the email we send. Linked approvers can also decide agents' requests in Slack.
- An approval covers the exact details shown. If anything changes, it has to be approved again.
- Moving money, deleting data and changing permissions always need a person.
See everything, stop anything
A live feed shows what every agent is doing. Each session has a timeline down to every model call and tool call, with its estimated cost.
- Spending caps per session, per agent and per account. The monthly caps for each agent and for the whole account alert you at 80% and 100%, and pause work when they're reached.
- Pause one agent, a whole company or the entire account at once.
- An audit log that can only be added to, chained so that any tampering shows.
100 integrations, plus your own
Agents use your apps through 100 built-in integrations, your own MCP servers and any API described with OpenAPI. Each action has a risk tier, and an agent only sees the tools its charter grants.
- Where an app gives you an API key or token from your account, that's the quick path. 45 of the 100 integrations, such as Gmail, Outlook Mail, Shopify and WhatsApp Business, connect through an app you create with the provider instead, so access stays in your hands.
- New MCP tools are reviewed before agents can use them, and a tool whose description changes needs approval again.
- Calls to addresses you add are checked first, so they can't reach private networks.
Knowledge and memory
Upload your procedures, FAQs, product notes and brand voice. Agents search them before they act, and you can limit any document to certain agents.
- Agents keep notes on facts, preferences and lessons learned. You can read and delete every note.
- What an agent reads in an email, ticket or web page is kept as information and never followed as an instruction.
Your AI models, your keys
Bring your own keys for OpenAI, Anthropic, Google Gemini, xAI, Mistral, Groq, DeepSeek, OpenRouter, Azure OpenAI, OpenAI-compatible endpoints. You pay your provider directly, at its prices, and Everdeck never marks them up.
- Keys are checked when you save them, and a failed check says exactly what's wrong.
- If a provider fails, an agent switches to its fallback model straight away.
- Estimated spend is recorded for every response, so you can see what each agent costs.
Security
Built to hold the keys to your business
Data in Singapore, every customer walled off from the others, credentials encrypted with keys held in AWS KMS, and two-factor sign-in for every person.
Risk tiers
Every tool carries one. Policies decide what needs approval.
- Tier 0ReadLook things up. No side effects.
- Tier 1InternalWrite to Everdeck's own records or notify your people.
- Tier 2ReversibleChange something outside that one step can undo.
- Tier 3ExternalAnything a customer or the public sees.
- Tier 4Money or irreversiblePayments, deletions, permission changes. Always a person.
Your company, always on deck.
Start a 14-day trial. Card first, nothing charged until day 14.