{AI Marketing Playbook}

← All examples

Ollama

Inference & servingBreakout · Show HN

Overview

Open-source tool that lets developers run open-weight LLMs locally with a single command (ollama run llama2), wrapping llama.cpp with a Docker-style model packaging format (the "Modelfile") and a one-line installer. Current scale: ~174,200 GitHub stars on ollama/ollama (June 15, 2026, GitHub API), 161,305 followers on @ollama (June 13, 2026, scraped). Written in Go, MIT-licensed. Bootstrapped: only a small pre-seed reported (figures vary by source, roughly $125K-500K incl. a Y Combinator note), no large VC round, which matters here because the growth was almost entirely organic, not paid.

First public appearance

Show HN on Hacker News, July 20, 2023, item 36802582, posted by founder Jeffrey Morgan (HN user jmorgan). Exact headline: "Show HN: Ollama, Run LLMs on your Mac" (rendered with the en-dash separator HN uses in titles). It linked directly to the GitHub repo, then github.com/jmorganca/ollama, not to a polished landing page. Key selling point led with: run large language models locally on your own Mac, zero cloud, built on top of llama.cpp, with a Docker-like packaging abstraction. Format: Show HN (the canonical dev-tool launch venue). The repo itself had been created June 26, 2023, and Wikipedia dates the initial public release to July 7, 2023, so the Show HN was the first real megaphone moment a few weeks in. The original site copy at ollama.ai (Wayback, Aug 2, 2023) read: "Get up and running with large language models, locally." with subhead "Run Llama 2 and other models on macOS. Customize and create your own." and "Available for macOS with Apple Silicon. Windows & Linux support coming soon."

Launch sequence

  • June 26, 2023
    GitHub repo jmorganca/ollama created (per GitHub API created_at). Quiet pre-launch build period.
  • ~July 7, 2023
    Initial public release (per Wikipedia). macOS-only, Apple Silicon.
  • July 20, 2023
    Show HN, "Run LLMs on your Mac" (item 36802582), 284 points, 87 comments. Founders Jeffrey Morgan (jmorgan) and Michael Chiang (mchiang) actively answered questions in-thread (memory requirements, the llama.cpp foundation, the Modelfile, the local API on port 11434). This is the ignition event.
  • August 1, 2023
    First ollama.com blog post, "Run Llama 2 uncensored locally." The blog started essentially on day one of the launch and became a steady model-availability content engine.
  • August 14, 2023 (3.5 weeks post-Show-HN)
    Repo at 5.1k stars, 280 forks (Wayback snapshot of the repo). README tagline at this point: "Run, create, and share large language models (LLMs)." The "Model library" table and "Tools using Ollama" (LangChain, Continue, LiteLLM, Raycast) sections were already present.
  • August 24 / September 9, 2023
    Blog posts "Run Code Llama locally" and "How to prompt Code Llama," timed to Meta's Code Llama release. Pattern established: ride every notable open-model drop with a same-week "run it locally with ollama run X" post.
  • September 26, 2023
    Second Show HN, "Ollama for Linux, Run LLMs with GPU Acceleration", 173 points, 58 comments. Expanded the addressable audience from Mac to Linux + NVIDIA GPUs.
  • October 6, 2023
    "Ollama is now available as an official Docker image", 197 points, 47 comments (submitted by a community member, not the founders). Reinforced the "Docker for LLMs" identity literally.
  • October 31, 2023
    Repo at 10.7k stars, 605 forks, 29 releases, 45 contributors (Wayback snapshot). README now led with the one-line installer curl https://ollama.ai/install.sh | sh, a large model-library table (Mistral, Llama 2, Code Llama, Vicuna, etc.), and a long "Community Integrations" list (LangChain, LlamaIndex, Continue, Obsidian, oterm, OllamaSharp, and ~15 more). The flywheel was visible.
  • January 25, 2024
    "Ollama releases Python and JavaScript Libraries", 607 points, 267 comments, the highest-scoring early Ollama HN story. Made Ollama trivially embeddable into apps, broadening from "tool you run" to "dependency you build on."
  • March 15, 2024
    "Ollama now supports AMD graphics cards", 633 points, 224 comments. Each hardware-coverage expansion was itself a front-page HN moment.
  • 2025-2026
    Product expansion into hosted cloud models, web search, and agent/IDE integrations (OpenAI-compatible API, Claude Code, Copilot CLI). March 2026: preview support for Apple's MLX. The cloud layer is the monetization path on top of the free local tool.

Channels & accounts

GitHub
github.com/ollama/ollama (originally jmorganca/ollama, now redirects), ~174.2k stars, ~16.6k forks (June 15, 2026). The primary owned channel and the de facto homepage for the dev audience.
X / Twitter
@ollama, 161,305 followers, account created Aug 7, 2023 (right after the Show HN), 8,292 posts, bio links to ollama.com and the repo. Founder accounts: Jeffrey Morgan @jmorgan.
Website
ollama.com (originally ollama.ai). Doubles as the model library / registry at ollama.com/library, which is the growth-flywheel surface: each model page is a discovery entry point with a copy-paste ollama run command.
Blog
ollama.com/blog, running since Aug 1, 2023, cadence tied to model releases and feature/hardware support. No RSS/newsletter surfaced.
Docs
in-repo README and docs/ (README itself served as primary documentation early on), plus the API docs.
Discord
discord.gg/ollama, linked from the README badge since the earliest snapshots (community Q&A, the "Discollama" bot lived in it). The server has grown to ~197,000 members (WebSearch, June 2026), a large community surface even though GitHub + HN + the model library were the primary growth engines.
Docker Hub
hub.docker.com/r/ollama/ollama, official image since Oct 2023.
No prominent owned YouTube, Telegram, or LinkedIn presence drove the early growth; the engine was GitHub + HN + the model library.

Amplification & KOLs

Hacker News community (organic, earned)
the single biggest amplifier. Multiple front-page stories (284, 173, 197, 607, 633 points), driven by the dev community resubmitting milestones, not by a paid push.
Open-source ecosystem integrations (earned, compounding)
LangChain, LlamaIndex, Continue (VS Code), LiteLLM, Raycast, Obsidian, Open WebUI, and dozens more added Ollama support and listed it in their docs. The README's "Community Integrations" list (15+ projects by Oct 2023) shows third parties pulling Ollama into their stacks, each becoming a referral source.
Simon Willison (named developer-influencer, organic/earned)
repeatedly recommends Ollama on his blog, e.g. "Ollama remains one of the easiest ways to run models on a laptop" (Jan 20, 2025, in his DeepSeek-R1 coverage) and use of ollama run phi4 (Jan 8, 2025). Indicative of how the practitioner-blogger crowd defaulted to Ollama as the recommended local-LLM runtime.
Model labs (indirect, earned)
Ollama's same-week "run X locally" posts for each major open-model release (Llama 2, Code Llama, Mistral, Phi, Gemma, DeepSeek) rode the attention of those launches. DeepSeek-R1 in Jan 2026 in particular drove a wave of "run it locally via Ollama" coverage.
No evidence of paid influencer campaigns or paid media; amplification reads as organic + earned throughout.

Traction inflection

The breakout was the July 20, 2023 Show HN ("Run LLMs on your Mac"), and the growth was then sustained and compounded by the one-line-install DX plus the model-library flywheel, with each subsequent platform/model expansion re-hitting the HN front page. Evidence: the star curve goes 0 -> 5.1k by Aug 14, 2023 (under four weeks after the Show HN) -> 10.7k by Oct 31, 2023 -> ~174k by mid-2026, with the steepest relative slope in the weeks immediately following the Show HN; that initial spike maps directly onto the 284-point front-page HN thread, and the founders' heavy in-thread engagement converted curiosity into installs. The reason it did not plateau like a typical Show HN bump: the product's DX made trial frictionless (single binary, curl | sh, ollama run llama2) and the model library turned every new open model into a fresh acquisition event, so the later 607-point (Python/JS libraries, Jan 2024) and 633-point (AMD, Mar 2024) HN stories each produced their own step-ups. Confidence: high for "Show HN was the ignition," based on the documented HN points, dated Wayback star checkpoints showing the post-Show-HN ramp, and the founder-engagement record in the thread. Confidence: high for "DX + model-library flywheel sustained it," based on README evolution (one-line installer, model table, integrations list growing across snapshots) and the recurring front-page milestone stories. The one thing not independently graphed here is a continuous day-by-day star curve (star-history.com is JS-only and would not fetch); the three Wayback checkpoints stand in for it.

Techniques & tactics

  • Launched on Show HN, the highest-signal venue for a dev tool, linking straight to the repo rather than a marketing site.
  • Founders personally worked the launch thread (answered memory/setup/architecture questions in real time), building trust and converting readers to users.
  • Radical DX as the marketing: single-command install (curl | sh), single-command run (ollama run llama2), zero config, single binary. The product demo is one line a reader can paste immediately.
  • Borrowed a known mental model: positioned as "Docker for LLMs" (the founders previously built Kitematic, an early Docker UI later acquired by Docker, which made the analogy authentic). The Modelfile, ollama pull/run/push, and an official Docker image made the analogy literal.
  • Built on llama.cpp rather than competing with it ("our goal was to build with/extend the project"), inheriting its performance and goodwill instead of fighting for it.
  • Model-library / registry as a discovery and re-acquisition flywheel: each new open model is a new landing page with a copy-paste command, and each major model release is a content + amplification moment.
  • README-as-marketing: the GitHub README carried the install commands, model table, quickstart, and a curated "Community Integrations" list that both helped users and signaled momentum.
  • Serialized milestone launches, each re-earning the HN front page: Linux + GPU (Sep 2023), official Docker image (Oct 2023), Python/JS libraries (Jan 2024), AMD support (Mar 2024).
  • Cultivated ecosystem integrations (LangChain, LlamaIndex, Continue, Open WebUI, Raycast, etc.) so other tools became distribution channels.
  • Stayed free and open (MIT) with no paywall on the core, maximizing adoption; monetization (cloud) layered on later, after the install base existed.

Sources