AI Open Source Radar — September 29, 2026

Collection basis: GitHub Trending dump bot issues. Today’s figures come from the Rain1er/trending issue for 2026-09-28 (updated 2026-09-28 18:31 UTC); the previous-day baseline comes from the earlier kouweizhu/agents-radar and happydog-intj/ai-xiaohongshu-daily issues (created 09-28 01:06 to 09:00 UTC). Repos with 40k+ total stars, long-running chart holders and non-AI repos are excluded.

Excluded today: vectorize-io/hindsight (+4,413, now past 40k at 40.7k), debpalash/VoiceStudio (+3,274, 43.2k), paperclipai/paperclip (+3,185, 92.3k), byoungd/up (64.6k), and PLFM_RADAR and coursebook (not AI).

mvschwarz/openrig — +781 ⭐ today (2 days in a row)

URL: https://github.com/mvschwarz/openrig
1,557 total stars · daily gain +114 → +781 (about +585% day over day) · total 990 → 1,557 · TypeScript · Apache-2.0

OpenRig is a multi-agent harness that runs several AI coding agents, such as Claude Code and Codex, together as one team. Rather than managing individual agents, it handles the operating infrastructure: which sessions are running, how they connect, and how they recover after a reboot. It is layered as a CLI/TUI/MCP server on top of a Hono HTTP daemon, domain services, and a foundation of SQLite, tmux and runtime adapters; the MCP server lets agents adjust their own topology.

Define agent topologies in YAML (RigSpec) with pods, edges and continuity policies
Boot a whole team with one rig up command, and snapshot or restore topologies by name
Topology explorer with graph and table views of rigs, pods and seats
Discover and adopt existing Claude Code and Codex sessions
Agent-to-agent messaging with rig send, rig broadcast and rig chatroom
Change a running topology with grow, shrink, launch and remove

Typical uses include implementation-plus-review pairs, product development with specialist agents, and agent-operated infrastructure (for example, a Vault specialist). As more teams run coding agents from different vendors side by side, this roughly 1k-star newcomer grew its daily gain nearly sevenfold in a day.

Practical use: Dev teams using both Claude Code and Codex can standardize implement-and-review agent teams as reproducible YAML, without a sprawl of tmux windows.

Tags: #MultiAgent #ClaudeCode #Codex #AgentOrchestration #tmux #MCP

dream-num/univer — +1,105 ⭐ today (2 days in a row)

URL: https://github.com/dream-num/univer
21,116 total stars · daily gain +895 → +1,105 (about +23% day over day) · TypeScript · Apache-2.0

Univer is an office SDK that handles spreadsheets, docs, slides, canvas, relational tables and PDF in one runtime, and it has recently repositioned itself as an ‘office harness for AI agents.’ It has a canvas-based rendering engine and a dedicated formula engine, and its isomorphic design runs the same way in the browser and in Node.js, so agents on a server can read, calculate and edit workbooks directly. Every capability ships as a plugin that can be added, replaced or lazy-loaded, and a Facade API exposes workbooks, ranges, formulas and documents through one consistent interface.

Canvas rendering plus a formula engine for large documents
Isomorphic browser/Node.js execution for server-side workbook automation
Plugin-first architecture with a preset mode
React 18+, Vue and Web Components support, with localization and theming
Comments, data validation, conditional formatting, drawings and hyperlinks
Real-time collaboration and import/export (Pro)

It is used to embed editors in SaaS and internal tools, in BI and data apps, and for agent-driven document automation. After slowing down following its September 23 spike, it has grown its daily gain again for a second day, a sign that demand for agents working directly on office files is holding. (Its only GitHub Trending #1 badge on Trendshift is from May 25, 2024; after a two-year gap, we don’t treat it as a long-running chart holder.)

Practical use: Teams that want agents to generate and check Excel-style reports can drive fully calculated workbooks headlessly from a Node.js server.

Tags: #OfficeSDK #Spreadsheet #AgentTools #FormulaEngine #TypeScript