Stop building your AI stack from 40 open tabs.
A curated map of open-source AI & agent tools. Every connection is a typed, human-reviewed relationship — with the why written down. How we curate →
495 repos · 24 topics · curated by ShipWithAI · new: impeccable, OpenDuet, auto-ontology →

Design vocabulary skill for coding agents: commands like polish, bolder, critique and audit, plus a 62-rule deterministic detector that catches AI-slop UI before it ships.

Proactive self-hosted AI assistant template on CopilotKit's OpenBot: morning briefs from mail, calendar and Slack, keeps working on its own computer, asks before sending, paying or deleting.

NVIDIA's ontology framework for agentic analytics: generate ontologies from metadata, ground agents in governed business meaning, validate generated SQL, and serve context via MCP, CLI and API.

Java cognitive memory engine for agents: working, episodic, semantic and procedural tiers with decay, consolidation and association graphs, SIMD hybrid recall, served over MCP, REST/gRPC and SDKs.

JetBrains' agentic IDE around the Pi coding agent: isolated worktree workspaces, a spec graph for spec-driven development, reusable skills and PR/CI flow — it learns your project as you work.
Spatial canvas workspace for coding agents: terminals, notes, git, browsers and devices on an infinite canvas, each task on its own worktree 'floor', with a Maestro terminal conducting the rest.
Higgsfield's official agent skills: image, video, 3D and audio generation across 30+ models, Soul identity training, product photoshoots, brand kits, thumbnails and faceless video.

Flight recorder for AI agents: policy-gates every tool call before it runs, holds risky ones for human approval, and signs each action into a tamper-evident log anyone can verify offline.

Rust LLM gateway in one binary: any OpenAI/Anthropic/Gemini client to any model, cross-vendor realtime voice, MCP tools inside chat calls, hedged requests and a live cost log.

Microsoft Research plugin that cuts coding-agent cost 18-31% on SWE-Bench by learning repo facts from past sessions, pruning unused tools and redirecting stuck failure loops — no agent changes.
Desktop agent orchestrator for running Claude Code, Codex and other coding agents in parallel, each in its own worktree — chat or terminal, diff review, git/PR flow, remote runtimes and a mobile app.

Self-hosted context layer and AI chat for teams: indexes 50+ workplace apps with permissions into a hybrid index, then serves agentic RAG, deep research, custom agents and an MCP server.
Mozilla.ai's self-hosted AI gateway: one OpenAI/Anthropic-compatible endpoint over 40+ providers via any-llm, with virtual keys, pre-dispatch budgets, routing policies and a usage dashboard.

Claude Code plugin that ties ticket → worktree → branch → ports → PR → CI → teardown together for parallel agent sessions, with a live dashboard pane, status line, toasts and a secrets vault.

Agent skill that art-directs coding agents into polished company websites — one design system, real photos and icons, restrained motion — verified in a real browser by a finish gate.

One Rust binary for agent search: BM25, vector and hybrid search over code, docs, logs and PDFs, an Elasticsearch-compatible API, MCP, and Query-1, which answers by running programs over facts.
Tencent's enterprise knowledge platform: one set of document bases served three ways, as RAG search, a reasoning agent with sandboxed skills, and a self-maintaining LLM wiki.

Tencent's team memory hub for agents: an API proxy turns conversations, docs and code into shared chat memory, skills, an LLM wiki and a code graph, reused across agents.

Meta's benchmark for open-ended bug hunting: given a real repository and no hints, an agent must find and fix as many bugs as it can. Scored as a micro-average of bugs resolved.

Local pre-push code review for coding agents: deterministic scanners (gitleaks, semgrep) plus separate Claude Code/Codex reviewer sessions, with a push gate. No account or key of its own.
The open-source terminal coding agent: TUI, desktop app and server, provider-agnostic across 75+ model providers including local models, with LSP integration and plugins.

Single Rust binary that turns JSONL, CSV, SQLite or any DB export into a ranked full-text index, giving agents short cited answer cards (~450 tokens) at any dataset size.
Independent auditor for AI agents: drops the agent into a simulation of your business, sends realistic requests, and scores operational alignment, permissions and red-team risk in ~2 minutes.
HelixDB's ontology builder: connect databases, tools and docs, let an agent draft an ontology, and load it all into HelixDB as one explorable, queryable graph with provenance.

Google Research's recursive harnessing framework: an agent refines its own harness (prompts, tools, memory, skills) per task via Propose–Probe–Compose, with model weights fixed.
One Rust program giving coding agents a code graph plus memory of docs, decisions and past sessions, served over MCP to Claude Code, Codex, Cursor and VS Code.

Agent memory plus a knowledge graph with ontology constraints in one SQLite file: pure Go, no server, no embedding model required. Plugins for Claude Code, Codex, dsh and Obsidian.

Agent skill that explains an idea by inventing a physical machine whose working is the concept, then builds it as an interactive isometric SVG, optionally with WebGL effects.

Read-only live monitor for Claude Code: a streaming graph of sessions, subagents, tokens, cost, file conflicts and alerts, as plugin, TUI and MCP server. Ships the AOS-1 spec.

Rust inference engine for on-device AI on Apple Silicon (Metal, unified memory) with Swift, Python and TS bindings, OpenAI-compatible server and a TypeSafe System One decision-model API.
macOS computer-use loop at ~$0.0002/step: OCR + accessibility tree read the screen, a TypeSafe classifier picks the next action, code clicks; an LLM writes only free text.
Hybrid GPU/CPU MoE engine that runs Qwen3.8-Flash-Next and DeepSeek V4.1 on consumer cards: hot experts cached in VRAM, rest computed in RAM, MTP drafting, local OpenAI/Anthropic API.

Reverse Engineer Anything: one MCP server plus CLI giving coding agents evidence-backed analysis of native binaries, JS/Electron apps, .NET assemblies and websites via Hopper, Ghidra or IDA.
Postgres extension for plain-language WHERE clauses: jev(row,'the customer is angry') filters, ranks and classifies rows via TypeSafe's Jev decision model. No embeddings, no vector column.

Context engine for coding agents: each session consults, then updates, a Markdown brain of specs, decisions, standards and lessons — private, team-shared via Git, and per-user. No embeddings.

Define a team of Claude Code, Codex and Pi agents in YAML and boot it with one command: persistent tmux-backed seats with roles, shared context and owned work, steered via a lead agent.
One email a week: what changed in the AI stack
New repos, new typed connections, the why behind each. No spam, unsubscribe anytime.