Autopilot Agents

Documentation Index

Fetch the complete documentation index at: https://docs.trysight.ai/llms.txt

Use this file to discover all available pages before exploring further.

​

Overview

Autopilot Agents are autonomous workers that act on your behalf. Where the rest of Sight AI surfaces opportunities for you to act on, Autopilot Agents take the next step — they read your opportunity queues, decide what to work on, and dispatch real actions (article generation, content refreshes, interlink insertions, site audits) on a schedule you control. Agents live at app.trysight.ai/agents.

Heads up: Autopilot Agents are different from the Article Generation Agents (Research Specialist, Content Writer, etc.) that power individual article generation. Those are roles inside the writing pipeline. Autopilot Agents are autonomous workers that operate the pipeline for you.

​

Meet the Agents

​

Article Creation Agent

What it does: Finds the highest-scoring Search Content Gap opportunities (queries with impressions where you have no matching article yet), checks for duplicates, and dispatches new articles.

• Picks opportunities above a configurable score threshold
• Calls siteHasArticleForKeyword before every dispatch to avoid duplicates
• Uses the suggested article type (listicle, how-to, explainer)
• Records a per-run summary so you can audit what it did

Default schedule: every 12 hours Default budget: 5 articles per run Use it for: automatically clearing your Content Gap backlog

​

AI Prompt Agent

What it does: Finds pending AI Visibility prompt opportunities — queries that LLMs answer where your brand isn’t mentioned — and dispatches articles shaped to be cited by AI answers (clear definitions, sourced lists, structured comparisons).

• Same dedup + dispatch flow as Article Creation, but for AI-side opportunities
• Uses a higher score threshold by default (since AI opportunities tend to need stronger signals)

Default schedule: every 12 hours Default budget: 5 articles per run Use it for: automatically improving your AI mention rate

​

Article Boost Agent

What it does: Refreshes pages that are gaining traction or underperforming. For each rising/refresh opportunity, it picks one of three approaches:

1. Freshness pass — bump dates, regenerate intro and meta if the article already covers the query well
2. Targeted section addition — insert one new section that answers the query, when there’s a clear gap (default approach)
3. Full refresh — replace the body wholesale, only when explicitly enabled and the article is stale

After refreshing, it republishes to your CMS. It never creates new articles — it only improves existing ones. Default schedule: every 12 hours Default budget: 3 boosts per run Use it for: automatically capitalizing on Rising Pages and Refresh opportunities

​

Interlinking Agent

What it does: Reads suggested internal-link opportunities (from the Interlinks tab) and inserts well-placed anchors into source articles.

• Pulls the source article body
• Decides where the suggested anchor text fits most naturally
• Prefers paragraphs already discussing related concepts; never forces the link into the intro
• Skips rows where there’s no natural fit (with a logged explanation) rather than placing a bad link

Default schedule: every 24 hours Default budget: 5 anchors per run Use it for: automatically building internal-link equity to underperforming articles

​

Site Performance Agent

What it does: Runs Lighthouse against your highest-traffic pages weekly and writes a prioritized report of fixes.

• Audits Performance, Accessibility, Best Practices, and SEO
• Flags missing JSON-LD on AI-generated articles
• Files a single report ranked by impact, not a flood of individual issues
• Notifies you with a one-paragraph TL;DR

Default schedule: weekly Use it for: keeping technical SEO healthy without having to remember to check

​

Backlink Outreach Agent (coming soon)

What it will do: Find editor or author emails for high-DR articles that mention your competitors, draft personalized pitches, send them, and track replies.

​

Listicle Outreach Agent (coming soon)

What it will do: Pitch existing roundup-style listicles (“Top 10 X tools”, “Best Y of 2026”) to add your brand to the list, then track replies.

​

How Agents Work

Every agent run follows the same pattern:

1. Pull a budget of high-scoring opportunities from the relevant queue
2. Dedup against existing articles, recent runs, and CMS state
3. Decide and act — call the configured tools (dispatch article, refresh article, insert anchor, etc.)
4. Record a per-run summary so you can audit every action

Each agent has a constrained tool surface — for example, the Interlinking Agent can read articles and insert anchors but cannot create new articles, by design.

​

Configuring an Agent

Open the Agents page, click into any agent card, and you’ll find:

• Status toggle — on/off
• Schedule — how often the agent runs (in hours)
• Score threshold — minimum opportunity score the agent will act on
• Action budget — maximum actions per run
• Tool-call budget — maximum total tool calls per run (a safety net)
• Model — which LLM the agent uses (defaults to Claude Haiku for simple agents, Sonnet for reasoning-heavy agents)
• Custom settings — agent-specific knobs (e.g., allow_full_refresh for the Boost Agent)

​

Activity Log

The detail page also shows:

• Recent runs — every run with its summary, action count, and tool calls
• Tool-call drilldown — click a run to see every tool call (input, output, latency)

This is the audit trail: you can see exactly which articles each agent created, which interlinks each agent inserted, and why each agent skipped a row.

​

How They Relate to the Rest of Sight AI

You can mix manual and autonomous workflows freely — accept some opportunities yourself in the UI, let agents work through the rest.

​

Safety & Cost Controls

Agents are conservative by default:

• Action budgets cap how many real actions any single run can take
• Tool-call budgets cap total LLM calls per run, so a misbehaving agent can’t run up a bill
• Schedule throttling prevents back-to-back runs
• Site eligibility checks prevent agents from running against unlicensed or paused sites
• Per-team article caps are enforced before any dispatch — agents can never push you past your plan limit
• Trial caps are enforced for trial teams before any dispatch

​

Best Practices

• Start with one agent — turn on Article Creation first, give it a week, then layer in the others
• Tune the score threshold to your tolerance — a high threshold means fewer, higher-confidence actions
• Pair Article Creation with Article Boost — Boost refreshes existing articles; Creation fills gaps. Together they cover most of your weekly SEO work
• Use Interlinking after a content burst — once you’ve published several articles, run Interlinking to weave them together
• Review the activity log weekly at first — once you trust the agent’s pattern, you can drop to monthly

​

Plan Availability

​

Next Steps

• Set up Search Opportunities so agents have a queue to work from
• Connect your CMS so refreshed and created articles can publish automatically
• Read about the Article Generation Agents (the in-pipeline roles agents call into)
• Compare with Autopilot mode — the older time-based scheduler