Ollama

Jeffrey (Jeff) Morgan and Michael Chiang founded Ollama in 2023 to make running open-weight language models on a laptop feel like using a package manager: `ollama run` downloads a model, applies a Modelfile/runtime configuration and exposes a consistent local API. The MIT-licensed project spread rapidly across macOS, Linux and Windows and became infrastructure behind editors, agent frameworks and tens of thousands of community integrations. Ollama does not train the underlying major models; it packages, quantizes, distributes and executes models from Meta, Google, DeepSeek, Qwen, Mistral, NVIDIA and others, subject to each model's own license. From 2025 it added optional cloud models so users can preserve the same local tools/API while offloading models too large for personal hardware; desktop apps, model sharing and agent launch flows broadened the product. Current individual pricing is Free ($0) and Pro ($20/month or $200/year), with Pro offering larger hosted models, three concurrent cloud models, 50x free usage and private-model upload/sharing; Team/Enterprise offerings are separately described/quoted. Local inference can keep prompts/data on the user's hardware, while cloud use is governed by Ollama's privacy/security terms and the company says cloud does not retain model inputs. In July 2026 Ollama announced a $65m Theory Ventures-led Series B, following a $15m Benchmark-led A; total funding is $88m, implying about $8m of seed/early capital. Investors include Benchmark, Theory, 8VC, YC, Garage, Pace, 49 Palms, GTMfund and prominent infrastructure founders. TechCrunch reported nearly nine million users/developers and a team of roughly 14 around the round. Risks include upstream model license/provenance, insecure exposure of local APIs, malicious model artifacts, resource/driver compatibility, cloud compute economics and competition from LM Studio, llama.cpp, vLLM, Docker Model Runner and hyperscaler/model-provider APIs.

Founding story

Morgan and Chiang saw open models becoming capable but difficult to install, quantize and operate, and created a Docker-like command-line experience that hid hardware/runtime complexity without forcing data into a hosted API.

Business model

Open-source local runtime with freemium hosted open-model cloud and paid individual/team/enterprise services.

Pro annual/monthly subscriptions, cloud compute/usage allocation and private model features; Team/Enterprise subscriptions/contracts, while local runtime and public models remain free.

Traction

July 2026 reporting: nearly 9m users/developers; company says cloud token volume has more than doubled monthly on average. Pricing page cites 40,000+ community integrations. Exact active/paid users, revenue and cloud gross margin are private.

Latest developments

Raised $65m Series B, scaled hosted access to large open models such as GLM, Nemotron, DeepSeek, Kimi and MiniMax, and integrated agent launch workflows while maintaining the free local runtime.