What people are building right now
Collections group the catalog by domain. Topics cut across it — each one gathers the analysed projects behind a trend that is actually driving work in 2026, with the demand signal stated up front.
Open-source AI agent frameworks and orchestration
Agent frameworks give you the loop — planning, tool calls, memory, retries — so you build the task instead of the plumbing. This is the deepest category in the catalog, which makes choosing harder than finding.
Open-source coding agents and tools for Claude Code, Cursor and Codex
Coding agents write, refactor and test code from a prompt. The open-source layer around them — skills, harnesses, review hooks, session tooling — is where most of the practical leverage now sits, and it is far less crowded than the assistants themselves.
Open-source MCP servers and tooling worth building on
The Model Context Protocol is how an assistant reaches your data and tools without a bespoke integration each time. Registries now index tens of thousands of servers, which moves the hard part from writing one to choosing which are worth trusting.
Open-source RAG, vector search and agent memory
Retrieval is what stops a model inventing answers about your data. These projects cover the whole path — chunking, embedding, storage, ranking and the memory an agent keeps between runs.
Local and self-hosted AI you can run on your own hardware
Open-weight models are now good enough that keeping inference in-house is a practical choice rather than a compromise. These projects cover local model runners, private assistants and the self-hosted infrastructure around them.
Open-source LLM inference, serving and optimisation
Inference tooling decides what a model costs to run and how fast it answers. These projects cover serving engines, batching, quantisation and the memory tricks that make a model fit hardware you already own.
Open-source security and code review tooling for AI-written code
AI writes code faster than teams can review it, and a measurable share of that code ships with exploitable flaws. These projects close the gap: scanners, review gates and analysis that run before a merge rather than after an incident.
Open-source browser automation, scraping and agent-driven browsing
Browser automation moved from testing into agent infrastructure: an agent that can open a page, read it and act on it needs a browser it can drive reliably and cheaply.
Open-source terminal and CLI tools worth stealing ideas from
The terminal became the main interface for AI-assisted work, and the tooling around it followed. These projects range from full TUI applications to single-purpose utilities small enough to read in an evening.
Open-source Rust projects to build on
Rust took over the performance-critical layer of developer tooling: terminals, parsers, package managers and inference runtimes. These projects are the ones with something reusable inside.
Open-source projects with the strongest commercial potential
These are the records scored 5 out of 5 for commercial potential — projects where a real problem, a payer and a plausible product all line up, not just a high star count.
Beginner-friendly open-source projects that are still worth building on
Projects rated easy to implement but still scored 4 or 5 for potential — the overlap between something you can get running this week and something that is actually worth running.