What Kautuka can do.
Use Kautuka to give agents tasks from your browser or a channel, set what they can access, review actions when needed, and follow the recorded results. It also supports repeatable workflows, scheduled runs, durable memory, and connected mail.
These capabilities are implemented in the current product. Feature availability can vary by installer, configuration, and connected provider; check the current install guide before you begin.
Delegate and review
- Run configured agents from browser chat and from Telegram or Slack.
- Choose which tools your agents can use and which actions need your approval, per tool, skill, or connector.
- Review requests, recorded actions, related agent work, and results.
- Use local or cloud model providers, with a capability check per saved model. Cloud models receive the prompts and context sent for their calls.
- Let one agent call another as a tool for self-contained work, with approval available on that delegation.
Repeatable work
- Connect agent tasks into a process, check results, and choose the next step using rules.
- Pause a process for a person to continue, ask for changes, or stop.
- Keep each run’s definition version and step progress, so eligible work can continue after a restart.
- Schedule work to begin later or repeat on a cron schedule, with pre-run approval when you want it.
Continuity and access
- Keep durable memory facts against an agent, separate from the session history they came from.
- Invite people by email, limit each account to selected agents, and support optional Google or Microsoft sign-in.
- Configure an organisation name and industry as default persona context.
Connected services
Connect Gmail or Microsoft 365 mail to find and read messages, review attachments, prepare drafts, and send mail. Sending begins with an “Ask” policy; each agent can use a different policy, so review its configured permissions. Other connectors can be added over HTTP, as a managed local command, as a declared container, or by importing an .mcpb bundle. Browser chat and a connected channel can both reach agents on your appliance, and supported file outputs can be returned through the connected channel.
Operate and observe
- A dashboard brings system and appliance health, recent sessions, and update state together.
- Usage shows consumption per run and summarised over time.
- Updates are checked against a signed release manifest.
Availability and limits
Version 0.1.20 installers are published for Linux x64 and the Windows appliance running through WSL. Linux agent chat requires Podman; the Windows installer provisions the WSL appliance. See the security guide for shared-agent scope, connector trust, sandbox boundaries, and recovery limits. Sharing a system agent shares its tools and configuration, not one person’s files or remembered facts; and not every interrupted external side effect can be safely repeated.
What this release does not include
- WhatsApp is deferred rather than rejected, and no unofficial transport is authorised. Telegram and Slack are supported.
- Permanent self-service deletion of a channel registration is not exposed; suspend and recovery are.
- macOS is not part of this release.
- Pricing for any paid offerings has not been announced.
Technical details
Workflow definitions have revisions, and each run retains the definition snapshot it started with. Checkpoints persist step progress. The platform recovers eligible active workflow instances after a Gateway restart and preserves waits for approval. This does not guarantee that every external side effect is safe to repeat or exactly once.
Mail connectors support listing/searching, reading, supported attachment retrieval, draft creation/update, and sending. Mail sending starts with “Ask,” but approval policy can be changed per agent. See Security for credential and container boundaries.