Open Source
Explore the latest AI open-source projects from GitHub and HuggingFace.
Explore the latest AI open-source projects from GitHub and HuggingFace.
FrontierAgent is an open-source agent runtime, terminal UI and evaluation harness published by ApodexAI, built for the company's own Apodex-1.1 models but designed as a provider-independent framework. The frontier-agent TUI ships two workflows: a stateful ReAct mode where one agent researches, reads files, writes deliverables and runs shell commands inside a task-scoped sandbox; and an Agent Team mode where a coordinator keeps a live task board, delegates bounded parallel work to sub-agents, collects structured reports, and synthesizes a final answer, optionally using a separate fast reporter for evidence review. Tools share one sandboxed filesystem (read-only /inputs, working /workspace, controlled /outputs for deliverables); mutating operations require diff approval unless run with --yes, and sessions are checkpointed, traced, and revertable via /revert. A new instruction typed mid-run is queued and injected at the next safe turn boundary rather than interrupting the active run; in Agent Team mode this steers the coordinator while already-dispatched sub-agents finish. Setup requires Git, Python 3.12, uv and an OpenAI-compatible model endpoint; pre-built Docker images and native install scripts cover macOS, Linux and GPU hosts, with Docker optional and local GPU serving via SGLang. The bundled evaluation harness runs each benchmark question in its own subprocess with resumable multi-run experiments and deterministic or model-based judges, supporting fourteen public benchmarks including BrowseComp, xbench-DeepResearch, Humanity's Last Exam (text-only), FrontierScience-Research, DeepSearchQA, WideSearch, FrontierSearchBench, GDPval and APEX; the README notes its GDPval implementation uses deterministic deliverable validation and intentionally excludes the agentic pairwise grader used elsewhere. Using this harness, ApodexAI reports its own Apodex-1.1 model scores higher on the Agent Team workflow than on ReAct across APEX-Agents, GDPval, FrontierFinance, FrontierScience-Research, BioMysteryBench and HLE, and that both configurations beat the company's prior Apodex-1.0 checkpoint on the same tasks. The project is Apache-2.0 licensed.
OpenClaw is an open-source, local-first AI gateway with 366K GitHub stars that routes AI responses through WhatsApp, Telegram, Slack, Discord, iMessage, Teams, and 15+ other platforms — zero cloud dependency.
OpenClaw
Open-source personal AI assistant connecting to 13+ messaging platforms with local gateway architecture, voice support, and multi-agent routing.