v0.1.0 - open source - MIT

A harness for AI agents that need to run.

Randal wraps AI coding agents in a persistent loop with memory, schedules, credentials, and real channels. Use OpenRouter by default, or bring Anthropic and OpenAI keys directly.

~/randalworkflow
active workflows
01 support ticket -> fix
02 morning brief
03 voice call beta
04 CI alert -> rollback
05 inbound lead
-> support ticket arrives in Discord
scout reads logs - builder opens fix - memory stores outcome
ok: pull request ready for review
what you get

Eight serious agent primitives, one harness.

Persistent memory

Past decisions, conversations, reviews, and outcomes stay searchable across restarts and deploys.

Schedules and cron

Run briefs, health checks, queue sweeps, and follow-ups on a clock without keeping a chat open.

Real channels

Discord and HTTP are first-class. Voice is in beta through LiveKit, Twilio, Deepgram, and ElevenLabs.

Dynamic posse dispatch

The orchestrator can choose roles, model fit, and how many ephemeral subagents to spawn for the task.

three ways to run

Local, service, or hosted Cloud.

way 01 - local

On your laptop

shellcopy
randal serve --posse home-base

Hack locally, talk through Discord or HTTP, and keep the same engine you can later run as a service.

way 02 - service

Long-running service

shellcopy
randal init && randal serve

Use one YAML config for agents, schedules, credentials, channels, and runtime behavior.

way 03 - hosted

Randal Cloud

shellcopy
$99/mo + usage

Use the hosted fast path with managed memory, pre-wired channels, and bring-your-own provider keys.