Back to blog

playbook · outreach

One outreach stack for LinkedIn, X and Reddit

Founders selling software live in the same three places as their buyers. Each one has closed or priced out its API. browser-memory turns all three into the same set of tools.

browser-memory running the LinkedIn flow end to end: find founders, enrich their profiles, then send two personalized invites. Real run.

You want your agent to find people who need what I sell and reach out, personally. The hard part was never the message. It’s that LinkedIn, X and Reddit each fight automation in their own way, and their APIs are either gone, gated, or absurdly expensive.

With browser-memory the moves are already saved tools. The same three steps repeat on every platform: find the intent, enrich it, act on it.

The pattern, three times

Find surfaces the signal: someone asking for a tool, comparing competitors, describing the exact problem you solve. Enrich pulls the context that makes a message land: a profile, a bio and recent posts, a full thread. Act is the outreach: a connection note, a reply, a comment. The agent writes each message from the context; the tool just sends it.

How each platform fights back

The interesting part is that none of them are reached the same way, and browser-memory learns whichever path works, once.

LinkedIn. Moved its people search to a server-driven UI with no clean API. The search tool reads the rendered results; profiles and connection requests still go through the internal Voyager API.

X. Locked its API behind an anti-bot token that can't be forged. Even the internal search endpoint returns 404 without it, so the tools read what the page actually renders.

Reddit. Still serves a clean JSON for any page. The easy one: search and threads come back as structured data straight away.

The nine tools

LinkedIn

readlinkedin-search-peopleFind people by role, company and location, with many searches in parallel.
readlinkedin-get-profileHeadline, about and experience, by API. No profile visit, no footprint.
writelinkedin-connectSend a connection request with a personalized note.

X

readx-search-tweetsFind tweets showing buying intent or competitor gripes.
readx-get-profileBio, follower count and recent tweets of the author.
writex-replyReply in the conversation, the native move on X.

Reddit

readreddit-search-postsFind threads where your product is relevant, across or inside subreddits.
readreddit-get-threadRead the post and its comments for context.
writereddit-commentAdd a value-first comment to the thread.

What a run looks like

“Find founders asking about cold-email tools and reach out.” The agent runs x-search-tweets for the intent, x-get-profile on each author, writes a reply that references what they actually said, and posts it with x-reply. Swap three tool names and the same run works on LinkedIn or Reddit.

No page is re-learned between runs. Each tool returns data, not a screenshot: a tweet is text, a profile is fields, a thread is a list. discover finds the tool, run executes it. Learned once, saved to a shared catalog, reused forever.

FAQ

Can an AI agent do outreach on LinkedIn, X and Reddit?

Yes. browser-memory gives an AI browsing agent pre-built skills for all three: find people or posts that show intent, enrich them with a profile or thread, and act with a connection request, reply or comment. The agent writes each message from the context; the skill sends it.

How does browser-memory reach platforms with closed or gated APIs?

Each platform is reached the way that actually works, learned once: LinkedIn's rendered search plus its internal Voyager API, X's page-rendered content where the API is locked behind an anti-bot token, and Reddit's clean JSON. The skill returns structured data either way — a tweet as text, a profile as fields, a thread as a list.

Does browser-memory send messages, or only find people?

Both, as separate skills. Read skills (search, get-profile, get-thread) find and enrich; write skills (linkedin-connect, x-reply, reddit-comment) act. You stay in control of what is sent, and the agent grounds every message in the context it just pulled.

Want this stack for your agent?

A shared memory of the web, so your agent reaches your buyers wherever they are, without re-learning the page every time.