The research window for this issue is May 4 to May 10, 2026. The clearest GitHub signal this week is not "another agent framework exists." It is that open-source attention is concentrating around three more concrete patterns: terminal coding agents, Claude/Codex-specific tooling, and MCP plus vertical agent workflows.

I do not treat stars as truth. Stars can reflect attention more than adoption. But when star growth, release velocity, repository content, and product direction all point to the same place, the signal is worth tracking.

1. Coding agents are the fastest-growing AI open-source category

OSSInsight's Trending AI Repositories snapshot in mid-May lists the top 28-day movers as anomalyco/opencode at +2.1k stars, anthropics/claude-code at +1.2k, and openai/codex at +940. All three are coding agents.

That is not a precise weekly measurement, but it shows direction: open-source AI attention is moving from generic agent frameworks toward coding agents that developers actually run in terminals and editors.

openai/codex describes itself as a lightweight coding agent that runs locally on your computer. Its GitHub page shows roughly 81.9k stars and 11.8k forks, with a 0.130.0 release on May 8. The README separates the CLI, editor integration, desktop app, and Codex Web paths. OpenAI is turning the coding agent into multiple entry points, not only a web task runner.

anthropics/claude-code is the Claude developer workflow entry point. Its GitHub page shows roughly 123k stars and 20.2k forks. The README describes a terminal agentic coding tool that understands your codebase, executes routine tasks, explains code, handles git workflows, and now includes plugins and custom commands/agents.

The competition is no longer only "who writes code better." It is who owns the developer workflow: terminal, IDE, git, review, plugins, memory, permissions, and internal distribution.

2. opencode shows that open-source coding agents need product surface

anomalyco/opencode ranks first in OSSInsight's 28-day top movers. Its README calls it "the open source coding agent" and offers both terminal usage and a desktop beta. The GitHub page shows 798 releases, with v1.14.48 listed as latest on May 11.

That matters because open-source coding-agent competition is not decided by models alone. Update cadence, installation, local and desktop experience, IDE/terminal entry points, and inspectability all affect adoption.

If Codex and Claude Code are official model-company entrances, opencode represents another path: a community-maintained operating surface that is not completely tied to one vendor's product shape.

3. Ruflo: Claude-centric orchestration breaks out

On May 5, duanyytop/agents-radar published an AI Open Source Trends issue naming ruvnet/ruflo as one of the day's strongest signals, reporting +2,598 stars and describing it as Claude-centric agent orchestration.

The GitHub repo now redirects to ruvnet/ruflo. Its page shows about 49.1k stars, 5.4k forks, 1,482 releases, and a May 5 Ruflo 3.6.30 release. The project describes itself as an agent orchestration platform for Claude, emphasizing multi-agent swarms, RAG, and Claude Code / Codex integration.

Do not take "swarm" language at face value. The more important point is that the repo maps to the same research problem discussed in the research weekly: as agents move from single tool calls to multi-agent coordination, open-source developers want an orchestration layer.

4. TradingAgents: financial vertical agents still pull attention

TauricResearch/TradingAgents is the other strong weekly signal. The May 5 agents-radar issue reported +2,182 stars that day; the GitHub repo currently shows roughly 73.8k stars and 14.4k forks. It is not new, but its May news notes a v0.2.5 release with a grounded Sentiment Analyst, GPT-5.5 model coverage, Qwen/GLM/MiniMax dual-region support, API-key auto-detection, remote Ollama support, non-US alpha benchmarks, and ticker path-traversal hardening.

Those changes are not just README decoration. They show a financial agent becoming more tool-like: more providers, more regions, better configuration, local/remote model support, non-US coverage, and security fixes.

I would still separate research tooling from investment performance. The README itself says the project is for research and is not financial advice. The real signal is that finance remains one of the strongest verticals for multi-agent workflow attention.

5. MCP and memory are becoming the interface layer

The same agents-radar issue highlights czlonkowski/n8n-mcp, zilliztech/claude-context, and mem0ai/mem0 in integration and knowledge categories. OSSInsight also lists modelcontextprotocol/servers among the leading MCP server repositories, still showing positive 28-day growth.

This is the open-source version of the enterprise control-plane problem. Agents cannot live on prompts alone. They need tools, workflow systems, company data, codebase context, and memory. MCP servers, memory layers, and code-search tools are the interface layer between chat and real systems.

What this GitHub track says this week

This week's open-source signal can be summarized in one sentence: agent open source is moving from abstract frameworks toward workflow entry points.

The most interesting projects are not necessarily the most general. They are the ones entering the terminal, codebase, Claude/Codex ecosystem, financial research workflow, and MCP integration layer.

That matches user behavior. Developers do not need another agent demo. They need agents inside the path where work already happens.

What I am watching next

First, whether opencode, Codex, and Claude Code keep widening the release and plugin surface: review, sandboxing, workspace memory, and organization distribution.

Second, whether Ruflo-style orchestration repos can absorb star growth through issues, PRs, documentation, and tests.

Third, whether vertical agents like TradingAgents can make benchmarks, backtests, data sources, and risk boundaries clearer. Finance is a good agent domain, but it is also the easiest place to confuse a demo with a money machine.

References

• OSSInsight, Trending AI Repositories on GitHub, accessed 2026-05-12
• GitHub, duanyytop/agents-radar issue #932: AI Open Source Trends 2026-05-05, 2026-05-05
• GitHub, openai/codex
• GitHub, anthropics/claude-code
• GitHub, anomalyco/opencode
• GitHub, ruvnet/ruflo
• GitHub, TauricResearch/TradingAgents