ContextBridge turns local models, private machines, VPSes and model APIs into one controlled pool. Apps describe what they need; CB routes jobs to a compatible worker based on capabilities, policy and live capacity. It supports self-hosting, explainable routing, E2EE, schedules, deterministic pipelines, MCP and bounded multi-step workflows, without forcing you onto a single runtime or cloud.
⏰ This product is scheduled for September 25th, 2026 12:01 AM PDT.
Upvoting is disabled until the launch is live.
Maker
📌
Hey Product Hunt 👋
I’m Angus.
I built ContextBridge because I got tired of rebuilding the same AI plumbing for every project, especially while I already had perfectly usable hardware sitting around.
My own setup is a mix of PCs, a rack, VPSes, shared hosting, local models and external APIs. I wanted applications to describe what they need without having to care too much about which machine, runtime or provider ends up doing the work.
That became ContextBridge.
CB turns those resources into one controlled pool. A PHP app on shared hosting can, for example, submit a job through a relay and have it executed by a model running on my PC. The scheduler filters workers by requirements such as task, model, tags, RAM/VRAM and policy, then ranks the compatible ones using current capacity. route explain shows why something was or wasn’t selected.
One of my more ridiculous test setups is an old 4 TB HDD with about 2.6 TB of models, Python tooling, RAG experiments and lots of smaller specialized models. Wiring setups like that together used to take me days. Now I can add them to the pool without rebuilding the integration every time.
ContextBridge also includes schedules, deterministic pipelines, MCP and a bounded planner for multi-step workflows. The planner can propose work, but it cannot give itself additional providers, budget, egress permissions or retries.
It is not meant to be an Ollama wrapper. Ollama is just one possible resource. The goal is a neutral layer where adding another machine, model runtime or API makes the pool more useful instead of creating another integration to maintain.
CB is written in Go, self-hostable and currently pre-1.0. It runs on Windows, Linux and macOS.
I use AI heavily for documentation, review, wording and parts of coding and testing, but changes still get reviewed against the actual code and tested in isolated or mirrored environments before I keep them.
At this point, I’d really like to see other people’s weird setups. If you find a bad assumption, confusing workflow, routing edge case or something that simply breaks, please tell me.
My rubber duck has stopped responding, so you’re up. 🦆