Skip to content

AI Agents & Presets

A curated gallery of installable agents — named presets over the capabilities Lithora already has. Install one, and it works your PRs, backlog, or releases in the background, always staging its actions for your approval.

AI Workspace
IDLE
issue #418github · open

users lose draft state on refresh; editor is slow on big docs; we should add autosave.

Waiting for work…

Agents reason over your linked context, propose a plan, wait for your confirmation, then execute — from a red build to an owned fix, or an issue to a merged PR.

What an agent is

An agent is not a separate AI product — it is a named preset over features Lithora already ships: automations, the AI Workspace, GitHub sync, and the memory graph. Browse and install them from Dashboard → Agents. There are two kinds:

Automation agents

Installing one instantiates a real automation template. “Installed” means your team now has a live automation created from that template — it runs on triggers like a PR opening or a nightly schedule.

Page agents

A feature that lives on its own page — Sprint Planner, Secret Sentinel, Workspace Q&A, Code Search, or Architect. “Install” just opens it; there is nothing running in the background.

The catalog — Code

GuardReview

On each PR, reads the diff, scores risk, and stages a review (comment or request-changes) for your one-tap approval.

CI-Triage Conductor

When CI fails, reads the logs, diagnoses the failure, and opens a fix task — storm-protected so a flapping pipeline can’t spam you.

PR Describer

When a PR opens, drafts a clear, reviewer-friendly description from the diff and stages it for approval.

Secret Sentinel

Deep-scans each PR’s diff for leaked credentials. Secrets are always redacted — only a fingerprint is stored.

The catalog — Plan

Backlog Triager

On a nightly schedule, reviews open work, proposes priorities and labels, and parks a digest for you to approve in the morning.

ExpertRoute

When an issue is opened, proposes the best assignee from code ownership and past work (staged for approval).

Standup Bot

Each weekday morning, summarizes what the team shipped and where work stands, and posts it in-app (and optionally Slack).

Sprint Planner

Pick a project and a capacity — proposes a coherent set of backlog tasks for the next sprint and creates a cycle from your selection.

The catalog — Ship & Know

ReleaseScribe

When a release is published, drafts grouped, human-readable release notes from the merged PRs and stages them for approval.

Workspace Q&A

Ask in plain language — answers from your team’s tasks, notes, and projects, and cites its sources.

Code Search

Index a linked repo and search its code in plain language — results deep-link to the file on GitHub.

Architect

Generate a Mermaid architecture diagram from a repo’s structure or a description.

The approval model

  • Staged, not autonomous: Agents that touch GitHub draft the change — a review, a description, release notes — and stage it. Nothing is posted to GitHub until you approve it.
  • Auditable: Each automation agent keeps a run history — you can see exactly what fired, when, and what it proposed.

Where agents run

Automation agents inherit the limits and quotas of the capability they wrap — for example, GuardReview draws on your plan's pooled PR-review quota, and every agent respects your team's role-based permissions.

Plan requirement

Agents sit on top of automations and Lithora AI, so they need an active Pro or Scale plan. Concretely: automation agents require the automations entitlement (Pro and Scale), AI PR reviews draw on the pooled per-seat quota (5 per seat on Pro, 15 on Scale), and agent conversations spend from the team's daily AI message pool (500/day on Pro, 3,000/day on Scale). See Automations and Lithora AI.