The top 100 GitHub repositories for AI agents and AI optimisation
The Gods Repo Library is a curated catalogue of 100 open-source GitHub repositories for building AI agents and optimising AI systems, sorted into 12 categories covering orchestration, coding agents, MCP and tooling, browser control, retrieval, vector stores, evaluation, prompt optimisation, inference serving, fine-tuning, data ingestion and reference material. Every entry carries a written note on what the project is for and when to reach for it, plus its licence, language, star count and last commit date read live from the GitHub API. It costs $9.99 once and arrives by email.
There is no shortage of lists like this. Most of them are a scrape of whatever was trending, dressed up with emoji, and half the links go somewhere that has not seen a commit since the year before last. The star counts are frozen at whenever the author last cared. That is the format this one is trying not to be.
What we sell here is the filtering. Over the past two years we have built agents for clients in healthcare, construction, hospitality and field services, and the expensive lesson each time was the same: picking the framework is the easy decision, and the one teams spend three weeks on. The hard part is everything underneath it. What you log. What happens on the fourth retry. Whether you can tell, at two in the morning, why the agent called the wrong tool. So this catalogue is weighted accordingly: the orchestration section is deliberately small, and the evaluation, observability and inference sections are deliberately not.
Half of it is not about agents at all, which is the other reason it exists. An agent that works in a demo and costs four dollars a run is not a product. The inference, quantisation and routing sections are the ones that turn a working prototype into something you can put a price on, and they are the ones that almost never appear on awesome-lists next to the agent frameworks.
Below you can see two entries from every category, for free, including the notes. That is 24 repositories in full. The other 76 are what the $9.99 buys.
One thing worth saying plainly, since you can check it yourself: all twelve repositories from our free Agent Library are in here too. They are the obvious ones, and a catalogue of 100 that left out LangGraph to feel exclusive would be a worse catalogue. So if you already have the free list, what you are buying is 88 repositories you have not seen from us, plus the deeper notes and live figures on the twelve you have.
- All 100 repositories with a written note on each, emailed to you
- 88 of them are beyond the free list above
- A permanent link to the live version, refreshed against the GitHub API
- No subscription and no account, just the library
Before you buy, the terms of this sale
- The full catalogue is delivered to the email address I give, immediately, the moment my payment clears.
- Because it is delivered in full straight away, all sales are final: no refunds, no partial refunds, no exceptions, for any reason.
- I am asking for that immediate delivery and I give up any cancellation or cooling-off period that would otherwise apply to digital content.
- If the email does not reach me, Frenchy Digital will re-send it to my address for as long as I need, and that re-send is the remedy rather than a refund.
- The repositories listed are third-party open-source projects. Their availability and their licences are theirs to change, and I will check any licence myself before using it commercially.
The same terms are in full in our terms and conditions, and you will be asked to accept them once more on the Stripe payment page.
You can change the address at checkout, and that is where it is delivered. Secure checkout by Stripe, and we never see your card.
What is in the 100
Every category, its size, and two entries from each shown in full so you can judge the notes before paying for the rest.
Orchestration: where the agent lives
19 repositoriesThe loop itself. Pick one. Picking two is how projects die.
- langgenius/dify156,909· TypeScript · See repository
A full platform with a visual builder over agentic workflows and RAG. The fastest route to letting non-engineers change the thing without a deploy.
Last commit 2026-09-23
- langchain-ai/langchain146,901· Python · MIT
The ecosystem, not just the framework: integrations for nearly every model and store. Worth it for the adapters even on projects that orchestrate elsewhere.
Last commit 2026-09-23
17 more in this category, each with its own note.
Coding agents you can point at your own repo
6 repositoriesAgents whose domain is software. The most battle-tested agents in existence, because their failures are visible in a diff.
- OpenHands/OpenHands88,907· TypeScript · MIT
A full development agent that edits, runs and tests inside a sandbox. The most complete open implementation of the pattern, and a good architecture to read.
Last commit 2026-09-23
- cline/cline69,095· TypeScript · Apache-2.0
An autonomous coding agent available as an IDE extension, CLI and SDK, with a plan-then-act loop and explicit approvals. Strong on keeping a human in the path.
Last commit 2026-09-23
4 more in this category, each with its own note.
MCP, tools and the integration layer
10 repositoriesAn agent with no tools is a chatbot. This is the plumbing that changes that, and MCP is now how most of it is wired.
- modelcontextprotocol/servers90,556· TypeScript · See repository
The reference MCP servers. Read two of them and you will understand the protocol better than any explainer will manage.
Last commit 2026-09-22
- upstash/context762,344· TypeScript · MIT
Feeds current library documentation to a coding agent, which is the cheapest fix for a model confidently using an API that was removed two versions ago.
Last commit 2026-09-23
8 more in this category, each with its own note.
Browser and computer control
5 repositoriesFor systems that expose no API. Brittle by nature, so budget for it breaking rather than pretending it will not.
- browser-use/browser-use115,981· Python · MIT
The default answer for pointing an agent at a website. Expect maintenance every time a target site redesigns.
Last commit 2026-09-18
- bytedance/UI-TARS-desktop39,094· TypeScript · Apache-2.0
A multimodal agent stack that operates a real desktop GUI. The clearest open look at where computer use is heading.
Last commit 2026-09-11
3 more in this category, each with its own note.
Retrieval: giving the model your data
9 repositoriesMost 'the model is dumb' complaints are retrieval problems wearing a costume.
- infiniflow/ragflow91,190· Go · Apache-2.0
Retrieval built around deep document understanding, with citations back to the source. Strong on the messy PDFs real clients actually send.
Last commit 2026-09-23
- run-llama/llama_index52,288· Python · MIT
The document-to-index toolkit with the widest set of readers and index types. Where to start when the data is varied and nothing else fits.
Last commit 2026-09-23
7 more in this category, each with its own note.
Memory and vector stores
7 repositoriesWhere the embeddings actually sit, and what remembers a conversation between sessions.
- mem0ai/mem065,854· Python · Apache-2.0
Drop-in memory for agents that extracts and updates facts across sessions rather than replaying whole transcripts. Check what it sends where before storing anything regulated.
Last commit 2026-09-22
- milvus-io/milvus46,223· Go · Apache-2.0
Built for scale from the start. Worth the operational weight at very large collections, and overkill well below that.
Last commit 2026-09-23
5 more in this category, each with its own note.
Evaluation, observability and guardrails
11 repositoriesThe boring half nobody posts about, and the half that decides whether you can ship.
- langfuse/langfuse34,950· TypeScript · See repository
Traces, evals, prompt management and cost tracking, self-hostable. The one thing here we install on day one of an agent build, not before launch.
Last commit 2026-09-22
- mlflow/mlflow28,108· Python · Apache-2.0
The ML platform that grew LLM tracing and evaluation. The path of least resistance for teams already tracking experiments in it.
Last commit 2026-09-23
9 more in this category, each with its own note.
Prompt and output optimisation
5 repositoriesStop hand-tuning strings. Compile them, constrain them, and test them like code.
- stanfordnlp/dspy38,218· Python · MIT
Declare what the module should do, let the optimiser compile the prompt and few-shot examples against a metric. The most important idea in this category.
Last commit 2026-09-23
- linshenkx/prompt-optimizer35,478· TypeScript · See repository
A practical tool for rewriting and comparing prompts side by side, usable by people who will never open an IDE. Good for handing prompt work to the team that owns the copy.
Last commit 2026-09-21
3 more in this category, each with its own note.
Inference, serving and routing
10 repositoriesThe AI optimisation half of this library: throughput, latency, and the bill at the end of the month.
- ollama/ollama181,498· Go · MIT
The friendliest way to run local models, with an OpenAI-compatible endpoint that lets you develop offline and switch providers by changing a base URL.
Last commit 2026-09-23
- ggml-org/llama.cpp129,246· C++ · MIT
CPU-first inference in C and C++ that runs almost anywhere, and the origin of GGUF. Still the reference for quantised local inference.
Last commit 2026-09-23
8 more in this category, each with its own note.
Fine-tuning, quantisation and compression
7 repositoriesWhen prompting has genuinely run out of road, and only then.
- unslothai/unsloth76,604· Python · Apache-2.0
Fine-tuning that is markedly faster and lighter on memory than a naive setup, with notebooks that run on a free Colab GPU. The easiest honest starting point.
Last commit 2026-09-23
- hiyouga/LlamaFactory74,986· Python · Apache-2.0
Unified fine-tuning across hundreds of models with a web UI and sane defaults. Removes most of the yak-shaving from a first fine-tune.
Last commit 2026-09-14
5 more in this category, each with its own note.
Getting the data in
5 repositoriesWeb pages, PDFs and codebases turned into something a model can read without choking.
- firecrawl/firecrawl183,496· TypeScript · AGPL-3.0
Turns sites into clean markdown for models, handling the JavaScript and the crawling. The fastest way to stop writing scrapers by hand.
Last commit 2026-09-23
- opendatalab/MinerU80,507· Python · See repository
High-quality PDF and Office extraction that keeps tables, formulas and reading order intact. The difference between usable retrieval and noise on technical documents.
Last commit 2026-09-22
3 more in this category, each with its own note.
Courses, cookbooks and reference
6 repositoriesNot libraries. The reading that stops you learning these lessons the expensive way.
- anthropics/claude-cookbooks52,915· Jupyter Notebook · MIT
Runnable notebooks from Anthropic covering tool use, retrieval and evaluation. Less theory, more code that runs on the first try.
Last commit 2026-09-22
- huggingface/agents-course32,784· MDX · Apache-2.0
A structured free course from fundamentals to a deployed agent, framework-agnostic enough to transfer. The best organised starting point here.
Last commit 2026-09-15
4 more in this category, each with its own note.
How the list is kept honest
The judgement and the facts are kept in separate files on purpose. Which repository belongs here, which category it sits in and what the note says are written by a person. The star count, fork count, language, licence, creation date and last commit date are never typed by hand: a build script reads them from the GitHub API, and the page shows the date it read them.
That script also enforces the two rules that keep a list like this from rotting. A repository that has been archived fails the build. So does one that has gone 180 days without a commit, with two documented exceptions for written references rather than libraries, and those two show their last commit date like everything else. If a project is renamed or moved, the build fails rather than quietly serving you a redirect.
Licences are reported exactly as GitHub resolves them, including the ones it cannot resolve, which show as “See repository”. We do not interpret them for you and we are not lawyers. Check the licence yourself before any of this ships commercially, particularly the AGPL entries.
One more thing said before you pay rather than after: the catalogue is delivered in full the moment the payment clears, so all sales are final and no refunds are offered. That is why two full entries from every category, the complete category list and the counts are on this page for nothing. Buy on the sample, not on hope. The full policy is in our terms and conditions.
Questions people actually ask
- All 100 repositories with a written note on each, emailed to you
- 88 of them are beyond the free list above
- A permanent link to the live version, refreshed against the GitHub API
- No subscription and no account, just the library
Before you buy, the terms of this sale
- The full catalogue is delivered to the email address I give, immediately, the moment my payment clears.
- Because it is delivered in full straight away, all sales are final: no refunds, no partial refunds, no exceptions, for any reason.
- I am asking for that immediate delivery and I give up any cancellation or cooling-off period that would otherwise apply to digital content.
- If the email does not reach me, Frenchy Digital will re-send it to my address for as long as I need, and that re-send is the remedy rather than a refund.
- The repositories listed are third-party open-source projects. Their availability and their licences are theirs to change, and I will check any licence myself before using it commercially.
The same terms are in full in our terms and conditions, and you will be asked to accept them once more on the Stripe payment page.
You can change the address at checkout, and that is where it is delivered. Secure checkout by Stripe, and we never see your card.
Already bought it and lost the email?
Not ready to pay for a list?
Fair. The free Agent Library covers twelve repositories across the same ground and costs an email address, nothing more.
Ready to Build Your Application?
Let's discuss your project. Our team will help you build solutions that transform your business.