llm-agents

OthmanAdi/planning-with-files

Persistent file-based planning for AI coding agents and long-running tasks. Crash-proof markdown plans, session recovery after /clear and compaction, per-turn re-injection against context rot, deterministic completion gate. Manus-style. Install from npm, the Claude Code plugin marketplace, or npx skills. Codex, Cursor, OpenCode, 60+ agents.

Fosowl/agenticSeek

A fully local, voice-enabled agent that browses and codes for you — AgenticSeek positions itself as a private, local alternative to Manus AI: it browses the web, writes and debugs code, and plans multi-step tasks using local reasoning models like DeepSeek-R1, with no cloud or subscription fees. Voice interaction is in progress. Useful for privacy-conscious developers wanting autonomous agents without API costs.

beenuar/AiSOC

A self-hostable SOC that lets AI agents triage alerts, auditably — AiSOC fuses alerts from your security tools, correlates them against 2,530 executable rules from a 7,015-rule library, and lets AI agents investigate each incident with every prompt and tool call logged for replay. It runs self-hosted with no API keys required, ships an MCP server for Claude, Cursor and Continue, and keeps a human approving any action before it reaches a vendor.

IBM/mcp-context-forge

A unified gateway that federates MCP, A2A, and REST/gRPC APIs — ContextForge is an open source registry and proxy from IBM that merges tools, agents, and APIs behind one endpoint, with centralized governance, discovery, and observability. It includes 40+ plugins, OpenTelemetry tracing, and runs as a compliant MCP server deployable via PyPI, Docker, or Kubernetes for multi-cluster scale.

samugit83/redamon

An autonomous AI pipeline from recon to merged patch — RedAmon chains recon, exploitation and post-exploitation into one pipeline, mapping the attack surface into a graph and attacking from a Kali sandbox. It triages findings, writes code fixes and opens PRs, with human approval gates, and works as both an MCP client and server for Claude Code or custom agents.

fudan-generative-vision/Remory

Soft memory that survives context compaction for LLM agents — Remory trains a learned soft memory, conditioned on an agent's compaction summary, so a frozen model can keep working across context resets. It targets long-running LLM agent sessions, with online training support (e.g. r-codex) letting small open models act as persistent research co-workers.

Continuum-AI-Corp/OrcaReplay

OrcaReplay — Time travel for AI agents. Record, replay, fork, and debug any agent run with any model. Built by the OrcaRouter.ai team.

uclanlp/OpenVLHarness

Official Code for OpenVLHarness: Real-World Visual Agentic Tool‑Use with Multimodal Memory

xingjian-bai/opal

An AI agent rewrites your PyTorch optimizer mid-training — OPAL wraps a PyTorch optimizer so a coding agent (Claude Code) can inspect training reports and rewrite the optimizer and analysis programs live, adjusting update rules, learning rates, momentum or clipping without touching model, data or loss. Until it intervenes, runs match the plain optimizer exactly. Aimed at researchers exploring agent-driven training dynamics.

OmniJev/Flide

Exact state forking speeds decision model inference 9.5x over vLLM — Flide is inference infrastructure built for Jev-style decision models. Its exact state forking lets it serve the Jev /v1/systemone API up to 9.5 times faster than vLLM and SGLang, while guaranteeing each answer matches what the model would give for that question alone. Useful for teams running LLM-based decision or judging pipelines who need both speed and deterministic output.

longsurf-ai/openchart

A free, open-source charting desktop app with AI agents built in — OpenChart is a free desktop trading workspace that pairs charts and technical analysis with AI agents for alerts, research, and opportunity discovery. It includes Tea, a Pine Script-like editor, and targets traders and quants who want LLM agents woven into their existing charting workflow, self-hostable.

1105216375-alt/resimind

Neuro-symbolic agents with verified memory. MCP math tools, an algebra repair Skill, and an open challenge. Verify solutions, learn checked rules, and reuse.

dukesun99/Corpus2Skill

Turning document corpora into navigable skill trees for agents — Official EMNLP 2026 Findings code for Corpus2Skill, which compiles enterprise document corpora into a hierarchy of Anthropic Skills. Agents navigate SKILL.md/INDEX.md files and fetch documents on demand at query time, replacing vector-search or BM25 serving. Useful for researchers and builders exploring alternatives to embedding-based RAG for enterprise QA.

davidalmeida90/glassbench

A common harness for TradingAgents, AI Hedge Fund and your own trading bots — Glassbench plugs multi-agent finance systems like TradingAgents, AI Hedge Fund and Vibe-Trading into one adapter, running each on the same stock and date and logging every agent run to a shared database. It adds a live front end, LLM/framework benchmarks, backtesting, and Interactive Brokers paper trading, useful for anyone comparing agent trading strategies head to head.