AI draft workflows

Turn synced artifacts into drafts on autopilot.

AI draft workflows watch your connected integrations and produce changelog drafts automatically. Start in draft-only mode, pick the artifacts that matter, and let the workflow run on a schedule or in response to new changes.

What AI draft workflows do

A workflow watches your synced integration artifacts and produces changelog draft entries automatically, either on a fixed schedule or in response to new changes.

Schedule-based

Run daily, weekly, monthly, or quarterly. Chainlog collects artifacts since the last run, deduplicates, and generates a draft.

Sync-triggered

Fires after an integration sync completes. Best when you want a draft every time new commits or issues arrive.

Artifact-count triggered

Fires when the number of changed artifacts since the last run crosses a threshold you set.

Step-by-step setup

Follow these steps to create your first AI draft workflow.

Connect an integration. Go to Integrations, connect GitHub, GitLab, Jira, or Linear, and sync at least one repository or project. The workflow needs synced artifacts to summarize.
Navigate to the AI area. Open Dashboard → AI. The AI Draft Workflows panel is at the bottom of the page.
Name the workflow. Pick a name that describes the audience or cadence, like “Weekly product update” or “Daily engineering digest”.
Choose the trigger type. Use “On a schedule” for most cases. Set the cadence (daily, weekly, monthly, or quarterly) and pick a first-run time.
Pick a delivery mode. Start with “Draft only” so you can review output before it reaches readers. Switch to “Auto-publish” only after quality is consistent.
Select a prompt preset. Use the platform default to start, or pick a saved preset from Settings if you have a specific voice or format.
Select artifacts (optional). Choose specific synced artifacts (repos, projects, boards) to feed the draft, or leave empty to include all synced artifacts automatically. Narrower scope produces more focused output.
Tune the guardrails. Set minimum artifacts (skip runs with too little activity), dedupe window (avoid summarizing the same changes twice), and minimum confidence score (filter low-quality generations).
Create the workflow. Click “Create Workflow”. The first run will execute at the time you set.

Configuration reference

Every field on the workflow form and what it controls.

Cadence

How often the workflow runs: daily, weekly, monthly, or quarterly. Only applies when trigger type is “On a schedule”.

Delivery mode

Draft only creates an unpublished entry for review. Auto-publish makes the entry live immediately, but requires a saved prompt preset marked eligible in Settings.

Min Artifacts

The minimum number of changed artifacts required for a run to proceed. Prevents drafts from low-activity periods. Default: 1.

Dedupe Window (h)

Hours to look back for previously summarized artifacts. Artifacts already included in a recent run are skipped. Default: 72 hours.

Min Confidence

Score between 0 and 1. Generations below this threshold are discarded. Leave empty to accept all output.

Best practices

Patterns that keep AI workflows reliable and useful.

Start with one workflow in draft-only mode targeting a single repo or project.
Review at least five runs before turning on auto-publish.
Keep artifact scope narrow. A workflow covering every integration produces generic summaries.
Use the dedupe window to prevent the same commits from appearing in consecutive digests.
If output quality drops, check whether the prompt preset or artifact scope changed recently.

Auto-publish requirements

Auto-publish requires a saved prompt preset marked as “eligible for auto-publish” in Settings → AI → Prompt Presets. The platform default preset is not eligible.