Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama
A Hacker News submission links to a blog post titled "Notes on migrating 35kb prompts away from Anthropic/OpenAI to Self-Hosted Ollama+opencode." The title indicates the author moved preprompts of roughly 35kb in size off Anthropic and OpenAI hosted models and onto a self-hosted setup built around Ollama and opencode. The post is characterized as notes on gotchas, meaning the author documented problems or surprises encountered during that migration rather than presenting a formal benchmark or product announcement. The provided source text consists of the title, the publication date of 2026-09-14, and the blog's embedded WordPress CSS variables and layout rules. It does not include the body of the post, so no specific model names, versions, hardware, token counts, latency figures, costs, or failure modes from the migration are available to report. What can be stated from the source is limited to the migration's endpoints: away from Anthropic and OpenAI, toward self-hosted Ollama plus opencode, with preprompts of about 35kb. Any further detail about which gotchas the author hit, how the prompts were adapted, or how the self-hosted results compared would require the full post text, which is not present here.
The post signals that developers are documenting practical friction when moving large preprompts from hosted Anthropic and OpenAI models to self-hosted Ollama and opencode.