AI Open Source Radar — July 12, 2026

Only repos that are breaking out right now. Ranking is driven by recent growth (weekly star velocity); repos that have plateaued or are just holding their rank are excluded. Yesterday’s picks — ai-job-search (daily +3.4k → +622) and system_prompts_leaks (weekly +7.6k → +7.1k) — have cooled and are dropped today.

stablyai/orca — +4,100 ⭐ this week (≈ 16,000 ⭐ total)

https://github.com/stablyai/orca

Orca is an ADE (Agent Development Environment) that runs a fleet of AI coding agents in parallel. It launches 20+ agents — Claude Code, Codex CLI, OpenCode, Pi, Grok and more — each in its own isolated git worktree, all tracked from a single interface. Instead of pouring the full context into one agent, you give the same prompt to several agents and compare the results side by side; it uses your own API subscriptions directly, with no per-seat pricing and no vendor lock-in, for unlimited parallelism. It ships on desktop (macOS, Windows, Linux) and mobile, and its appeal as one of the few open-source tools built squarely for parallel-agent workflows is behind the recent surge.

Multi-agent orchestration: run Codex, ClaudeCode, OpenCode or Pi side by side, each in its own worktree, tracked in one place.
Design Mode: click a UI element in a real Chromium window to inject its HTML, CSS and a cropped screenshot straight into the agent’s prompt.
Mobile support: monitor agents from your phone and send follow-ups, with a notification when an agent finishes.
API flexibility: bring your own Anthropic, OpenAI, Google or xAI keys — no seat licensing, no vendor lock-in.

Practical use: stand up a “best-of-N” coding pipeline — run the same task across five agents in parallel and pick the best result — using nothing but your team’s existing subscriptions.

#ParallelAgents #AgentOrchestration #CodingAgents #GitWorktree #ADE #MobileAgentControl

diegosouzapw/OmniRoute — +4,100 ⭐ this week (≈ 15,000 ⭐ total)

https://github.com/diegosouzapw/OmniRoute

OmniRoute is a free AI gateway that unifies 231+ providers (50+ of them free) behind a single endpoint. It connects 24+ coding agents — Claude Code, Codex, Cursor, Cline, Copilot, Antigravity and more — through one config, and pointing any tool at /v1 automatically translates requests across OpenAI ↔ Claude ↔ Gemini ↔ the Responses API. Its pitch of “keep coding on free models without a paid subscription” is spreading fast among developers and driving the surge. Its core features are below.

RTK+Caveman stacked compression cuts tokens by 15–95%.
Millisecond-scale smart auto-fallback — when one provider’s quota runs out, the next takes over with zero downtime.
Aggregates 40+ free-tier pools (500+ models) into ~1.6B tokens/month (up to ~2.1B in the first month with signup credits), shown live on the dashboard.
MCP/A2A, multimodal APIs, and Desktop/PWA support.

Practical use: consolidate multiple paid LLM accounts behind one gateway with cost caps and auto-fallback to forecast and cut team-wide token spend.

#AIGateway #ProviderRouting #TokenCompression #AutoFallback #FreeTier #MCP #A2A

ogulcancelik/herdr — +4,000 ⭐ this week (≈ 15,000 ⭐ total)

https://github.com/ogulcancelik/herdr

herdr is an “agent multiplexer” that lives in your terminal. Written in Rust and shipped as a single binary with no external dependencies, it adds, on top of an ordinary terminal multiplexer (like tmux), awareness of the state of each running AI agent — the key differentiator. As demand grows for running several coding agents from a single control-room-style view, it has held a top trending spot and keeps climbing. Its main features are below.

A sidebar shows whether each agent is blocked, working, done, or idle via color-coded indicators.
Detach and agents keep running; reattach from any terminal or over ssh, and sessions survive restarts.
Both tmux-style prefix keys and mouse click/drag/split are first-class.
Supports many coding agents — pi, claude, codex, copilot, cursor, devin, droid, kimi, opencode, kilo, hermes, qodercli.

Practical use: remotely supervise long-running multi-agent jobs over ssh and see at a glance which agent is waiting for input, cutting idle wait time.

#TerminalMultiplexer #RustLang #AgentStateTracking #SessionPersistence #TmuxAlternative #CodingAgents

Source: findarepo — AI Agents (updated 2026-07-11), GitHub. Metric: weekly (7-day) star velocity.