Everything your agents need to do real work.
33 features in 7 areas, all in one workspace.
Your company of agents
Organise agents like a real team: roles, managers, and a CEO you can talk to.
Org chart
Teams, managers and agents on a live org chart. See who is running, waiting or idle at a glance.
Agents with a role
Every agent gets a role, instructions, a model, tools and guardrails. Changes are saved as versions you can roll back.
Agents that hire agents
Give your CEO the manage-company permission and say "make me a researcher agent". It builds out the org for you.
Sub-agents
Agents spawn helpers for parallel work and collect their results before reporting back.
Multiple companies
Run separate companies side by side. Agents, keys, budgets, memory and audit log are never shared between them.
Members & roles
Invite your team with role-based access, per company.
Work that runs on its own
Hand off work once. Agents pick it up on a schedule, on a trigger, or when another agent asks.
Tasks
Create tasks for any agent, or let managers delegate them. Follow every step in the run detail.
Triggers & cron
Run agents on a schedule or when something happens. One agent finishing can start the next.
Company chat
Talk to your CEO or any agent. Agents post their results where your team reads them. Attach screenshots and log files.
Message a running agent
Send a message while an agent is working. It reaches the agent mid-run through the run inbox, without restarting.
Recorded browser flows
Record a procedure once in Chrome. Agents replay it with integrations first, and only click through a browser as a last resort.
Search everything
One ⌘K palette for agents, runs, tasks and settings — or ask an agent straight from it.
Coding agents
Claude Code and Codex, running in their own sandbox, on the subscription you already pay for.
Claude Code in a sandbox
Coding agents use Claude Code with its own tools — read, edit, grep, bash, sub-agents — inside an isolated container.
Codex with your ChatGPT plan
Sign in with your ChatGPT subscription, or use an API key. Limits, cancellation and the kill switch keep working.
Automatic fallback
When Claude hits a limit, the run hands over to Codex (and back), and tells you once.
Git workspaces
Agents clone your repo, run installs, builds and tests, commit, and open a pull or merge request on GitHub or GitLab.
Knowledge & memory
Agents get better every run, and work from what your company already knows.
Memory that learns
After every run, agents reflect and keep what is worth remembering — within a daily budget. Say "remember…" to store something directly.
Knowledge base
Shared knowledge for all agents, searchable with semantic search. Connect an Obsidian vault.
Skills
Reusable instruction bundles in the open SKILL.md format. Import and export the same zips Claude and Codex use.
Safety & control
Autonomy with a leash. You decide what agents may do on their own.
Approvals
Policies per tool: allow, ask or deny. Risky actions wait in "Needs you" until someone decides — also from Telegram.
Isolated sandboxes
Every agent works in its own sandbox. Nothing runs on shared machines with access to other companies.
Network allowlist
Sandboxes only reach approved hosts. When an agent needs a new one, it asks and a human decides.
Budgets & kill switch
Spending limits per agent and company stop runs before they overspend. One Halt button stops everything.
Audit log
Every change, approval and tool call is logged, per company.
Encrypted secrets
API keys and tokens are envelope-encrypted at rest and never shown again after you save them.
Models & integrations
Any model, any tool. Agents reach you wherever you are.
Any model provider
Anthropic, OpenAI, xAI, OpenRouter, Ollama and any OpenAI-compatible endpoint. Bring your own keys.
MCP servers per agent
Give each agent exactly the MCP servers and tools it needs — nothing more.
MCP endpoint & tokens
Drive Mandistry from Claude, Codex or your own scripts with personal access tokens.
Notifications everywhere
In-app, web push, Telegram, e-mail, Slack or Discord. Approve or deny straight from Telegram.
Cost & insight
Know what every agent did, what it cost, and what went wrong.
Token & cost tracking
Every token and every cent, per agent, per run and per day. Subscription runs are tracked too.
Daily run analysis
Each day, errors are grouped per agent, tool and trigger, with a likely cause and a concrete fix proposal.
Run detail & log export
Replay a run step by step, see status at a glance, and download full diagnostics.
Help center & support chat
Tutorials in English and Dutch, with search, AI summaries and a support chatbot inside the app.