Custom agents

AI Research Wiki 2026: Auto-Building Knowledge Base with Deep Citation Syntheses

One robot, many AI brains. A Physical AI platform that controls an Adeept PiCar-B2 through swappable AI control interfaces behind a shared hardware abstraction. Built today: an LLM agent (Claude vision + tool use). Planned: VLAs and world-model / action-conditioned prediction.

Open LLM CLI - LLM Automation

端到端中文法律咨询 Agent：自研轻量调度框架、法条 RAG 与公网案例双源检索、NDJSON 流式 Web UI、自修复链路与运行指标观测

Easy-to-use, customizable, high-quality threat modeling skills for AI agents.

Deep-research LLM agent + RAG platform for internal bank-product audit — cited reports, PDF export, pgvector

AdvPP - AdvPL/TLPP Compiler and IDE with complete language support

AI-powered OSINT & security reconnaissance toolkit for Python — subdomain finder, WAF detector, JWT analyzer, email breach & SPF/DMARC checker, GitHub secret scanner + a natural-language LLM agent. For bug bounty & pentesting.

Discord meeting bot: local Whisper transcription, LLM summarization, and an agent that tracks tasks/decisions as searchable memory across meetings.

Small Python Agent runtime with planning, memory, tools, state machine, and multi-agent review.

Chrome-tillägg som låter en LLM sköta kundtjänstchatten åt dig — du pekar ut fälten, den skriver.

Template file system mcp server

Satellite-native crop yield intelligence: Spatio-Temporal Transformer + live Sentinel-2/ERA5 via Google Earth Engine, FastAPI backend, and an LLM tool-calling agent (Groq/Llama 3.3) with guardrails and automated eval. Research prototype — synthetic training data, documented known issues.

一个 7×24 小时不间断运行，帮你跑实验、用 Git 维护公开仓库、更新 LaTeX（Overleaf）论文的科研自动化工作流。实测效率提升 25 倍。Claude Code Skill，一条命令装进任意项目。

A 股多智能体投研编排系统：7 个专职 subagent 以文件契约协作，长文档上下文隔离，数字 claim 双源核验，公司与行业双研究链路

Claude Code / Codex skill for authoring and reviewing SuperDialog voice-agent playbook YAMLs — schema reference, 15-point failure-pattern checklist, interrupt-guard code pattern

A quantamental equity screener combining investment criteria selection with real-time market screening faciliated through LLMs

Zero-dependency LLM agent runtime — tool use, MCP, sandboxed self-written tools, checkpoint/resume, approval gates, bi-temporal fact memory, code-level guards (context bounds, loop guard, egress control, shadow mode), declarative tool policy, OpenTelemetry. Not a framework. 885 tests, CI Linux/Windows, 31 runnable demos, online visual builder.

企业级 RAG 知识库问答系统 + AIOps 运维诊断 Agent(Plan-Execute-Replan / LangGraph / MCP)

Dynamic Knowledge Graphs with Memory Decay for LLM Agents

Portable semantic memory layer for LLM agents over MCP

Deterministic, inspectable workflows for coding agents. Kilin is a local, CLI-first workflow runtime: describe a workflow in plain files, validate it before anything runs, then execute it through Codex, Claude Code, or OpenCode with explicit approvals and persisted runs.

LibreControl — one universal /remote-control for every coding agent you run (Claude Code, Codex, Gemini, Cursor, Copilot, any MCP tool), across all your machines, centralized into the single chat app you choose: Telegram, WhatsApp, Slack, Signal, Discord. Self-hosted, with approvals and audit.

agentic geotech engineer

Self-hosted, governed-autonomy SRE platform: an LLM agent that autonomously remediates infrastructure incidents behind a fail-closed prediction gate, mechanical verdicts, and a tamper-evident ledger. The agent can't act on a belief it hasn't checked.

Best AI Code Architect 2026: Diagram-First Agent Orchestrator with AST Editing & Atomic Commits

Local-first, layered memory kit for AI agents — stdlib-only, zero deps

Give your AI agent a phone — a thin Python CLI over the BlueBubbles REST API to read, search, send, react, and attach in iMessage, with draft-then-confirm send gating and an agent-facing SKILL.md.

Many models. One harness. Apache 2.0 runtime for multi-protocol agent orchestration: one goal in, a verified multi-agent run out across MCP, A2A, and Computer-Use.

Offline plan drift scoring for LLM agent traces

AI agents built spaCy model recommenders, then a fresh AI peer-review panel rejected all three

Local LLM-powered assistant for sysadmins and SREs. Runs 100% on your GPU.

Post-response grounding check + cross-family adversarial review for LLM chat agents

Локальный модуль машинного зрения для LLM-агентов на чистом Rust: метрический 3D scene-graph каждый кадр, MCP-сервер, $0/CPU/приватно

Claude Code Workflow that executes cys:plan implementation plans with independent tasks running in parallel via a dependency DAG — ships the cys plugin (design, plan, run, check, ship).

Runtime-neutral protocol for growing adaptive marketing agents: role + playbook + GEB learning, with schema validators and CI gates enforcing it. Pure spec — bring your own runtime (Codex, Claude Code, MCP, etc.).

LEASH-8: an 8-domain control model for AI agents with delegated authority. Scorecard, approval-design checklist, plan-vs-authorize pattern. Patterns from a live production agent operation, sanitized. Free, MIT.

163 specialist Claude Code agents + 24 routing skills + 16 strategy playbooks. Korean/Japanese Business Navigators, game-dev (Unity/Unreal/Godot/Roblox/Blender), XR/spatial, paid-media routing. skill-routing-arbitrator disambiguates the ~500-skill ecosystem. Fork of msitarzewski/agency-agents.

TeleAgent AI is an enterprise-grade, agentic omnichannel consumer intelligence engine engineered specifically for Deutsche Telekom AG (Germany & Europe) under Problem Statement 5: Omnichannel Consumer AI Engine for Digital Commerce.

Official MCP server for Proxidize — manage mobile & residential proxies from Claude, Cursor, and any MCP client. Rotate IPs, monitor usage.

GSD lifecycle orchestration as a native OpenClaw plugin — enforced research→plan→execute→verify→ship, hybrid retrieval, 0 Discord slots

Multi-agent collaboration engine with persistent, searchable, self-compacting memory. Agents research, critique and adjudicate over a shared SQLite memory with hybrid BM25 + embedding recall. Zero-dependency core, runs locally on Ollama.

Engineering architecture for runtime state modulation in Large Language Models.

Survivability-first quantitative research system. An AI council debates every architecture decision before code; deterministic, tested strategies do the trading. Walk-forward + purged CV + deflated Sharpe. LLMs never place trades.

AI-SOC — detects, explains and contains cyberattacks on critical infrastructure. Isolation Forest detection + RAG/MITRE attribution + policy-gated response, wired end-to-end. ET AI Hackathon 2026, PS7.

Rich Brain / Clean Hands: a two-brain delegation pattern for LLM agents that keeps the deciding brain's context clean and exiles heavy work to an isolated executor.

More human oversight can make an AI agent less safe. Headroom is a human-in-the-loop firewall for coding agents that measures when to trust the human: oversight as resource allocation, not just classification.

An agentic buyer's agent for an unrepresented Bay Area home purchase — deterministic Python core, LLM at the edges.

Action-graded severity scoring (L0-L6) for tool-using AI agents, computed from red-team execution traces.

Security & runtime checklist for moving LLM agents from demo to production: HITL gates, idempotency, prompt injection boundaries, tool risk classification (CC BY 4.0).

Decision governance for code and the agents editing it — track PRD→ADR→SPEC decisions, map them to files, and gate changes against them. Deterministic, honest CLI + MCP.

Deterministic execution harness for AI agents — goals in, structured tasks out, verified results back.

Infer least-privilege MCP tool policies from agent traces

Context Transform Engine.

ai built in os mainly on rust with some hint of c c++ assembly

Graph RAG on SQLite for AI agents: vector retrieval + hand-curated wikilink graph + cross-encoder rerank, with a zero-token per-turn memory ledger. Working pilot.

DualityCert: verifier-gated LLM repair of Seiberg-duality claims in QFT (arXiv:2607.23614). First instance of the QFTCert program.

ETH Zurich thesis project: a nine-agent AI pipeline for quant research, from hypothesis to backtest report

Judgment continuity for AI agents — load your project's scars, taste, and non-negotiables into every session. Judgment-layer companion to edda. (formerly project-doctrine)

Financial Intelligence Workstation — A native, cross‑platform desktop terminal with 37 AI agents, 100+ data connectors, 16 broker integrations, QuantLib suite, visual node editor, and embedded Python analytics. Free and open‑source (MIT). Your thinking is the only limit.

AI-powered code review and autonomous development platform built with TypeScript, React, FastAPI, PostgreSQL and LLM-based agents.

Privacy-first AI assistant that runs entirely on your machine. Local Ollama inference, embedded SQLite/Kuzu/ChromaDB, Rust-backed Brain Firewall, MCP connectors.

Production 5-agent pipeline automating pre-sales: discovery → scoping → pricing → proposal. 6 BUs, 22 services. Groq + Llama 3.3. Adversarial CriticAgent.

Kubernetes-native control plane for multi-tenant agent platforms on EKS. A Platform CR is the tenant boundary: namespace, IAM identity, KMS grants, scoped datastore access. Budget, SLO and eval policies are CRs. Bedrock for models, Envoy AI Gateway for the data plane.

Make any coding agent work like a frontier model. Drop-in Agent Skills for disciplined planning, evidence-first debugging, and live-system safety — plus a Python project scaffold with an agent contract, review checklist, and definition of done. Model-agnostic, zero dependencies.

BRAIN — The Mirror That Doesn't Lie | Local-first behavioral AI engine

Autonomous AI artist with mood system, episodic/semantic memory, ReAct thinking, and LoRA style training

Autonomous verifiable on-chain trading agent for BNB Chain: commit-reveal predictions sealed before the outcome, a drawdown circuit breaker that halts on breach, drawdown-budgeted Kelly sizing fit on 181k labeled outcomes, and an x402 machine-payable signal feed.

Autonomous AI agent that bids on and completes coding jobs on the NEAR AI marketplace to earn NEAR tokens - engineering case study

Auto-rollback for AI agent config changes. Zero dependencies, single Python file.

Self-hosted autonomous coding agent with executable skills and memory that survives sessions

🎮 AI dispatch agent that builds teams. Knows every agent's hexagon panel, tracks bonds, dispatches with personality. From the game Dispatch.

Framework-agnostic orchestration layer for AI agents: circuit breakers, multi-LLM council, workflow FSM, sweeps. Companion to Engraphis.

Quality-gated granularity control for multi-agent LLM pipelines

Give AI agents hands without handing them the keys: exact one-use approvals and verifiable receipts for registered actions.

Multi-agent orchestration with self-evolution, cognitive memory, and governance.

Production-ready multi-agent orchestrator — confidence-gated routing with a pluggable classifier, MCP or in-process agent dispatch, per-agent circuit breakers, and pluggable session/breaker persistence (Firestore, Postgres, Redis).

A reasoning-first idle engine for persistent LLM agents. Behaviour emerges from the agent reasoning about its idle state, gated by a cheap model-free safety layer.

Self-hosted Agent Ops — run, govern, observe, and improve fleets of AI agents from one local-first control plane.

Keep your agent and workflow. Add visibility first, limits when needed, and enforceable contracts for high-risk work.

Official documentation for Agentflow, built with Docusaurus and structured via the Divio system: tutorials, how-to guides, reference, and concepts.

The Kubernetes for AI Agents. An open-source orchestration control plane with deterministic DAG execution, policy governance, and A2A/MCP protocol support.

Build persistent digital lifeforms with local M4 brain, OpenClaw agent, Qwen core, and Git-based memory.

Sanctum — local-first agent orchestration engine. Cyclic state graphs, checkpointing, streaming, and robust tool-calling for local LLMs. Zero-dependency core.

A compact Python SDK for strict tool contracts, observable streaming events, and scoped local file capabilities.

A minimal, bus-driven Python framework for LLM agent orchestration that decouples communication, versioned prompts, and multi-agent patterns—helping AI architects build scalable, vendor-neutral agent systems without infrastructure lock-in.

Core engine for the autonomous development factory platform with workflow orchestration

Governed multi-agent orchestration: Contract · Artifact · Scope · Authority. Agents exchange artifacts, not chat transcripts.

A cognitive runtime architecture that transforms experience into reusable structure, reducing the reasoning required to operate over time.

Architecture blueprint for accountable AI agents — a cognitive kernel, fail-closed policy hooks, swappable model providers, and evidence-graded outcomes.

CodegniPy is a groundbreaking Python library that elevates AI to a first-class citizen of the language. It introduces a cognitive computing engine where deterministic code and non‑deterministic LLM reasoning coexist seamlessly, enabling you to write Python with intent rather than just instructions.

Multi-agent orchestration framework — design AI teams in YAML, route tasks to cheap or expensive models by role

Loop + Wiki agent framework — the wiki is the agent's mind. Identity, memory & capabilities live in plain markdown, loaded via progressive disclosure. Multi-agent message bus, global memory with feedback-driven pattern learning, REPL + web chat.

Multi-Agent AI Orchestrator 2026 🚀 | YAML, 6+ LLM Providers, ReAct & Swarm

Turn your vault into a private chat workspace by connecting Obsidian to an Ollama model

Perpetual, self-correcting development loop for Claude Code: an RSI backlog loop + MAKER-style per-step error correction (decomposition + multi-agent voting, with your tests as the judge).

Upload a bundle of files and get a shareable web URL. One static Go binary is both the server and the client, built for agents that need to hand a rendered page to a person.