Tell
Give Randal the outcome, not a perfect prompt.
Start in chat, Discord, HTTP, or a schedule. Randal turns the request into a bounded run with checkpoints and visible next steps.
Features
The pieces an AI agent needs
The pieces an AI agent needs
to keep working.
Randal is not just a chat surface. It is an operating loop for useful AI work: memory, schedules, scoped credentials, channels, dynamic workers, and checkpoints that keep long tasks reviewable.
Useful defaults
- Loops and checkpoints reduce context rot.
- Memory carries decisions across sessions.
- Schedules and channels make agents operational.
- Workers and provider choice keep tasks right-sized.
Tell / Trust / See
The operating loop in three moves.
Trust
Keep review gates and config boundaries explicit.
Source-controlled config, scoped credentials, visible memory, and stop/context controls keep the agent understandable while it works.
See
Watch the transcript, events, and outcomes accumulate.
Run output, job history, memory notes, and scheduler events stay available so future work starts warmer than the last run.
Core benefits
What changes when Randal is in the loop.
01 / Loop discipline
Checkpoints keep work achievable.
Randal splits long-running agent work into loops with clear checkpoints, so progress can be reviewed, resumed, or stopped before context gets messy.
02 / Memory
Sessions do not have to start cold.
Project memory keeps useful decisions, outcomes, run notes, and lessons searchable across future sessions without replacing source control or human review.
03 / Real channels
Work can arrive where teams already talk.
Discord and HTTP are practical paths today. Voice is beta, and the adapter model gives teams a place to connect their own surfaces carefully.
04 / Schedules
Recurring work gets a clock.
Cron jobs, one-shot reminders, and heartbeat wake items can run briefings, queue sweeps, checks, and follow-ups without keeping a browser session open.
05 / Dynamic workers
Specialists handle bounded slices.
The conductor can dispatch scoped workers for research, implementation, debugging, or review while keeping the overall loop small enough to understand.
06 / Model flexibility
Use the right provider path for the task.
Bring your own model/provider keys and choose the model path that fits the work. Randal does not need model billing bundled into the product to be useful.
07 / Scoped credentials
Access can match the job.
Credentials can be scoped by service and task so agents get the access they need for a run, not a blanket promise that every action should be automatic.
08 / Hosted or self-hosted
Choose who operates the runtime.
Self-host Randal as free OSS when you want control. Use Randal Cloud at $29 per seat per month when you want a hosted single Randal. BYO model keys stay direct, with no inference markup.
Objections
Plain answers before you wire an agent into work.
Is this just another chat UI?
No. Chat is the easiest way to start work, but the product value is the loop: sessions, checkpoints, memory, schedules, credentials, workers, and reviewable outcomes.
Does Randal connect every SaaS tool today?
No. This milestone keeps integrations truthful: current channel paths and built-in Tavily, Image, Video, and Notion connectors are shown; unsupported cards stay hidden.
Can I self-host without losing core features?
Yes. Local/self-host is first-class. Cloud is an operated path for teams that want the same core loop without maintaining the runtime themselves.
Operating paths
Hosted Cloud and self-host are product paths, not different promises.
Choose the operations model that fits the team. The core loop stays the same.
Self-host
$0
Run the open-source harness yourself, operate the dependencies, and bring your own provider keys.
Deployment docsRandal Cloud
$29 / seat / month
Use the hosted path when you want Hassion Studio to operate one always-on Randal for you.
See CloudStudio
$199 / seat / month
For posse-style multi-Randal workflows, setup help, support, and SSO-style team needs.
ContactNext step
Start with the path you can operate confidently.
Self-host for free/OSS control, or use Cloud for a hosted always-on Randal at $29 per seat per month. Model keys remain BYO, and Randal does not mark up inference.