AI assistant

goloom has a built-in AI assistant: a chat agent that lives inside the app and can actually do things for your team — look up data, draft and edit posts, set up campaigns and automations, and schedule publishing. It is driven by the same tool catalog as the MCP server, so an agent behaves consistently whether it runs inside goloom or connects from the outside. The difference is that the in-app assistant is conversational and screen-aware, and it asks for confirmation before any action that publishes, schedules, deletes, or auto-publishes.

Open it from the AI Assistant panel in the app. The conversation is always scoped to the team you are working in.

What it can do

Answer questions and surface insights — without leaving the chat:

your calendar, posts and a content search (get_calendar, find_free_slot, get_posts, search_posts)
connected accounts and their character limits (get_platforms)
engagement analytics and top posts (get_analytics)
follower growth over time and a single-metric history chart (get_account_growth, get_metric_history)
best-performing hashtags and the best times to post (get_hashtag_performance, get_analytics_timeslots)
your team’s brand voice, which it recalls before writing copy so drafts sound like you (get_brand_profile)

Draft and edit content — these run immediately (nothing is published):

save a new draft with per-platform variants (draft_post)
edit an existing draft or scheduled post in place — never a duplicate (modify_post)
define a reusable campaign format (create_campaign)

Schedule and automate — these are proposed and run only after you confirm (see Confirmation step):

schedule a post for a specific time (schedule_post)
delete a draft or scheduled post (delete_post)
set up a recurring auto-publishing automation (create_recurring)
set up an RSS automation that turns feed items into posts (create_rss_feed)

Screen awareness

The assistant can see what you are currently looking at, so you can talk in context. Say “shorten this post” or “schedule this for Friday” and it reads your active screen and the focused entity (get_current_view) instead of guessing.

When you have the composer open, the assistant edits the unsaved draft you are writing rather than creating a new one: it returns a suggested revision you apply with Apply in composer — default text and/or per-account overrides, nothing is saved until you do. You can detach the composer context, and use the per-account scope chips to include or exclude individual accounts from an edit.

Confirmation step

Reads, drafts and in-place edits happen right away. Anything that publishes, schedules, deletes, or sets up auto-publishing is different: the assistant does not execute it. Instead it shows a confirmation card (“Confirm: …”) with Confirm and Dismiss. The action only runs when you click Confirm.

Your click is not a trust shortcut — team access and the required token scope are re-checked when the action actually runs, exactly as for any other write.

Referencing things with @-mentions

Type @ to reference a specific account, campaign, recurring automation or RSS automation. The assistant loads that entity’s details into the conversation, so it works with the real thing instead of a guess. Mentioning an account also scopes the request to it — e.g. it targets that account in a new post, or sets only that account’s override when revising — and leaves the others alone unless you ask otherwise.

Slash commands and links

Type / for quick commands:

Command	Does
`/draft`	Start a prompt to draft a post.
`/campaign`	Start a prompt to create a campaign.
`/recurring`	Start a prompt for a recurring automation.
`/rss`	Start a prompt for an RSS automation.
`/clear`	Start a new conversation (clears the current thread).

Paste a link in your message and the assistant fetches the page’s readable text (fetch_url) so it can draft about real content — it won’t invent what a page says.

Safety

AI-produced drafts can flow through the review queue so a human approves before anything publishes.
Every write the assistant performs is recorded in the team audit log, attributed to the user who ran it.