Skip to main content
By the end of this guide a workflow posts a short digest of the pull requests merged in a repository to a Slack channel every weekday at 9am. A GitHub search finds the pull requests, an agent writes the summary, and the Slack integration posts it. None of those steps touches files or a shell, so the workflow runs as a no-machine run: it fires on Reliant’s servers even when your laptop is shut.

Before you begin

  • The reliant CLI, v1.21 or later, signed in (Sign in and tokens), and a Reliant project for the repository the workflow lives in.
  • The Slack and GitHub integrations connected under Settings → Integrations.
  • In an organization with other members: an organization AI account for the agent step. See What this does not do.
1

Invite the bot and find the channel ID

In the channel, run /invite @Reliant. The bot can post only where it is a member. Steps take the channel’s ID, such as C0123ABCD, not its #name. Slack shows the ID at the bottom of the channel’s details, and the slack/conversations.list@1 action lists the channels the bot can see with their IDs.
2

Write the workflow

.reliant/workflows/daily-slack-digest.yaml
since and the edges are plain CEL, evaluated on the worker. On a day with nothing merged, the run posts one line from quiet and never starts the agent, so it spends no tokens. tools: [] gives the agent no tools at all: it only writes. That’s what lets it run without a machine, because the stock agent’s default tools include the shell. requires: makes a run fail before its first step, naming the missing connection, instead of halfway through.
3

Test it with scenarios

Scenarios mock the GitHub search, the agent and the Slack post, so nothing is sent. One covers the quiet day:
.reliant/workflows/daily-slack-digest/scenarios/quiet_day_skips_the_agent.yaml
Add one for a day with merges (mock merged with two items, write with response_text, and post) and one where the post fails, then run them:
terminal
A failed Slack call doesn’t fail the run on its own: it sets is_error. The bot_not_in_channel scenario mocks post with is_error: true and checks that the run ends at failed, so a missed digest shows up in your Inbox. See Test workflows with scenarios.
4

Enable it with no machine

Commit the workflow. In the app, open Workflows → Library, open daily-slack-digest, choose Enable, set Runs on to No machine, and fill in repo and channel. From a terminal:
terminal
If anything in the workflow needs a machine, enabling is refused and the reason names it, for example needs a machine (...), so it cannot run with no machine. reliant trigger create can’t make this automation: its schedules always need a machine.
5

Verify the first digest

Fire it once instead of waiting for 9am, then look at the firing:
terminal
The newest row should say launched with a chat ID. reliant workflow status <chat-id> should show Status: COMPLETED, and the digest appears in the channel, posted by the Reliant bot. A run that ended at failed says why in its last message, under Workflows → Runs. The activation’s name comes from the workflow and trigger names unless you pass --name-prefix; reliant trigger list shows it.

What this does not do

  • In an organization with other members, the agent spends only the organization’s AI accounts. Automations there are created org_only. The run starts either way, and without an organization account that can serve the model it pauses at write. Quiet days never reach the agent.
  • Slack posts go through your connection. There is no separate bot identity to choose, as there is for GitHub; the message comes from the Reliant bot user of the workspace you connected.
  • It reads titles, not code. The agent sees each pull request’s title, link and author. Give it more, such as github/pr.get@1 results, by adding steps before write. The search returns one page, here up to 50 pull requests.
  • Firings missed during an outage are dropped, not replayed, once they are older than the catch-up window (10 minutes by default).

Troubleshooting

Run /invite @Reliant in the channel. The run ends at failed with this text.
channel must be an ID (C…, G… or D…) that the bot can see, not a #name.
Nothing in the Enable dialog asks for a Slack connection, because the trigger is a schedule. Connect Slack under Settings → Integrations for the account that owns the automation.

Next steps

Build a Slack bot that answers in threads

Start a run when someone mentions the bot, and reply in the thread.

Nightly dependency sweep

A scheduled workflow with no model that files GitHub issues.

Agent on a schedule

Cron, timezones, overlap and the catch-up window.

How automations fire

AI funding, no-machine runs and what each firing records.