Agents
Routines are Claude/Ollama-powered agents that run on a schedule with tools, budgets, and a sandbox. This section is the operator's manual.
A routine is one TOML file. It names an agent, sets a schedule,
declares the tools the agent may call (read-globs and shell commands,
both whitelist-enforced), points at an output destination, and caps
the budget in tokens per run and dollars per day. vinos-routine enable <name> generates a systemd user timer.
You’ll spend most of your time in three commands:
vinos-routine list— see what’s registered and when it runs next.vinos-routine run <name>— fire one ad-hoc.vinos-brief— open today’s outputs in a walker panel.
If you’re deep in it, jump to the routine spec for the authoritative TOML schema. Everything here is the operator’s view.
Running a routine
List, enable, run, log, and inspect cost for routines. Every command with real captured output.
ReadWriting your own routine
Full anatomy of a routine TOML file — schedule, agent, tools, output, budget. Copy from the shipped day-brief and mutate.
ReadProject routines
Commit .vinos/routines.yaml next to your code. Teammates run vinos-routine load . after clone and inherit the same agent set.
ReadTools and the sandbox
read: globs and shell: commands are whitelist-enforced by the runtime — not the model. Every shell tool runs inside bwrap.
ReadThe 80/20 router
route = auto runs local Ollama first, escalates to Claude only when needed. The cost math and the escalation policy.
ReadTroubleshooting routines
API key not found, Ollama unreachable, sandbox refusal, budget exceeded — every failure the runtime knows about, and its fix.
Read