Ideas for your next build
The AI projects worth a closer look. What they do, what you could build, and what to watch out for. Less directory digging. More getting somewhere.
Selected for usefulness. Not ranked by lifetime stars.
The useful part of the changelog
Three source-linked changes with a practical interpretation.
Editorial assessments are pinned to a specific release, not generated from a version number.
New structured decision endpoints and expanded KV cache options create new integration and memory trade-offs to evaluate.
Benchmark a representative workload before upgrading. New serving options are not universal speed guarantees.
The shortlist
An opinionated starting point, not an exhaustive ranking.
Sources checked . Not installation tests.
Give an agent more than a chat loop.
Planning, files, context management and delegation in one extensible harness. A useful starting point when a task takes more than one tool call.
A research assistant that gathers evidence and writes a report.
Know first Constrain tool permissions and require approval for consequential actions.
Choose the coding agent. Keep your model choice.
An open-source coding agent for working in a real repository. Worth exploring if you want to choose your provider rather than commit to one model ecosystem.
A small feature in an existing app, with tests and a reviewable diff.
Know first Review generated changes. Model usage and hosted services can still cost money.
Your PDFs deserve better than copy and paste.
Convert documents into structured representations for search, retrieval and downstream AI. Start here before building another chatbot on messy source files.
A searchable document library with structure and page references.
Know first Check extraction quality on your own layouts, tables and scans.
Turn a website into something AI can actually read.
A self-hostable crawler that turns web pages into Markdown and structured data. Useful when the hard part of your AI workflow is getting usable source material.
A research pipeline that collects public pages with their source URLs.
Know first Respect site terms, robots policies and rate limits. Do not bypass access controls.
Make open models your own.
Run and fine-tune supported models with a local interface and training tools. A practical route from using a model to adapting one to a specific task.
A fine-tuning experiment with a small, carefully checked training set.
Know first This release is a beta. Check GPU memory, model licences and training data rights.
Go from a model on disk to a model people can use.
Serve supported open models through an API, with batching and memory-conscious inference. Explore it when your project needs a shared model endpoint.
An internal model API, benchmarked against your actual workload.
Know first You own hardware, authentication, monitoring and capacity planning.
Leave with something working
Runnable scripts, sample inputs and explicit checks.
Document and research samples were executed. Local AI has client contract tests, not a model benchmark.
Extract three invoice fields with a page number and exact supporting text. Missing or conflicting values stay unresolved instead of becoming guesses.
Open the mini-build →Actual upstream runCollect two pinned release records into a compact evidence brief. Preserve the release URL and source excerpt before anyone writes a summary.
Open the mini-build →Contract tested; inference not runSend a synthetic extraction task to your own loopback Ollama server, then reject any response that does not match the expected answer.
Open the mini-build →Go one step beyond browsing
Try Docling on a short, non-sensitive PDF. Get Markdown you can inspect before feeding it into a search or AI workflow.
sample.pdf.convert.py.from docling.document_converter import DocumentConverter
converted = DocumentConverter().convert("sample.pdf")
print(converted.document.export_to_markdown())python -m pip install docling
python convert.py > sample.mdYour check: a readable sample.md that agrees with the PDF. Missing or scrambled content is a reason to investigate, not to trust the output.
Source-reviewed recipe, not tested on your hardware. Initial setup may download model assets. Follow the project’s installation and privacy guidance.
Follow the work
Published releases from our picks, newest first.
Upstream metadata refreshes on request, cached up to 15 minutes per server instance. Editorial review is separate.
New upstream version. Read the release notes; editorial impact review is pending.
New upstream version. Read the release notes; editorial impact review is pending.
New upstream version. Read the release notes; editorial impact review is pending.
New upstream version. Read the release notes; editorial impact review is pending.
Adds structured decision endpoints and expands low-precision KV cache support.
A versioned crawler release with Python and Docker installation paths.
Release dates come from the linked GitHub release pages. We do not use catalogue import dates as launch dates, or call star counts “trending”. Inspect the metadata receipts ↗
Keep exploring
Looking for something specific? Dig into the index.
Catalogue records are not endorsements or installation tests.
24 projects on this page
Clear filters →Free, read-only GPU waste scanner for Kubernetes inference clusters, by Paralleliq. Finds idle GPUs, oversized GPU tiers and wasted capacity in vLLM deployments.
Claude Code harness that refreshes its own context instead of compacting: at a threshold it writes a verified capsule, runs /clear, and resumes in a fresh session. Memory is plain
Catch your AI agents when they lie about what they shipped — verifies claims against git instead of believing the agent.
Agentic coding in the terminal: the model writes its own multi-agent workflows (Gigacode). Provider-agnostic, local-first, 15+ model providers, MCP support.
Practical course about Large Language Models.
The context layer for production AI agents: local-first context planning, compression and runtime for LLM apps.
Pause your AI agent. Ask a human. Resume with their answer. Open source human-in-the-loop (HITL) library for production LLM agents: Slack, email, and web dashboard. Typed Pydantic
Security-posture-first list of sandboxes for running AI coding agents' code, ranked by isolation, egress control and secrets handling
Extract structured data from semi-structured documents using any OpenAI-compatible LLM, with per-field grounding and confidence.
An AI SRE for Kubernetes: connects to your cluster, investigates with real tools, explains the root cause in plain English, and executes the fix only after you approve it. Peer-rev
Regression testing for LLM applications — with the baseline in your repository, not on someone's server.
Track LLM token usage by user, session and model without a metric label for every user. Prometheus or any OTLP metrics backend.
Metamorphic testing for AI outputs — find wrong LLM answers without an answer key
Jebadiah, an open System One decision model: trainer, data builders, evals and every run record
Private, self-hosted, no-code AI agent builder. Design agents on a visual canvas; run them on built-in local inference (llama.cpp, MLX, vLLM) or any OpenAI-compatible API. MCP tool
Open-source AI, ranked: apps, agents, models and the tools around them, each with strengths, weaknesses and a score from real GitHub signals. Re-checked twice a day.
A circuit breaker for your LLM API bill — hard budget caps + per-feature cost attribution for OpenAI & Anthropic.
PyPI capacity-planning CLI for LLM deployment. pip install chimeraforge.
Open-source self-hosted control plane for focused AI agents: roles, tasks, approvals, knowledge context, activity and cost visibility. Built by Felhen.
Evidence-first LLM-as-judge scoring for AI artifacts — papers, PRs, prompts, cold emails: synthesizes a rubric, collects quoted-evidence citations, scores only against that evidenc
Evaluation tools for Jev-compatible decision models, covering accuracy, calibration, hidden instructions, padding and consistency.
Build container images with purpose. Self-hosted OCI image builder with a web UI, plus a terminal client (TUI and CLI) for jumphosts and scripts.
Compile AI agent skills into cheap, fast, deterministic pipelines that run without an LLM in the loop
Claude Code mod: picks the model and reasoning effort of every subagent
Repository activity and recently indexed are different signals. Indexing an established project does not make it new.