1
Hermes Agent — Ecosystem & Related Tools Catalog
(reddthat.com)
Hermes Agent — Ecosystem & Related Tools Catalog
Date: 2026-08-09 Compiled from relatedrepos.com browsing of the tools installed on this box (mem0, codegraph, open-second-brain, rtk) + GitHub API verification. Categorized and sorted by fit for this Hermes Agent setup (Rust + Svelte, FicHub, Linux Mint).
Reality check first: Several names from past AI guides that looked like "hallucinations" (
headroom,ponytail,caveman) are actually REAL community repos — they're just NOT Hermes first-party tools. They exist as Claude Code skills / standalone CLIs that are Hermes-compatible via MCP or skills. They are listed below with their real purpose.
1. Code Intelligence & Code Graphing
Sorted by fit for local, offline repo traversal (FicHub = Rust + Svelte + TS).
| # | Tool | Stars | Lang | Why / Fit |
|---|---|---|---|---|
| 1 | CodeGraph (colbymchenry/codegraph) — INSTALLED |
~15k | TS | The current best local MCP code intelligence. Pre-indexed SQLite + tree-sitter call graph, single codegraph_explore tool = low token overhead. Already indexed FicHub (299 files). Gold standard for blast-radius analysis. |
| 2 | GitNexus (abhigyanpatwari/GitNexus) |
45k | TS | Zero-server, browser-based knowledge graph creator. Good semantic understanding of Rust traits / Svelte props. Heavier (browser UI) but strong for deep code understanding. |
| 3 | Graphify (Graphify-Labs/graphify) |
105k | Python | Turns codebase + docs + SQL schemas + PDFs into queryable knowledge graphs. /graphify skill for Claude/Cursor/Codex. Strong for architecture queries; heavier on tokens than CodeGraph. |
| 4 | Understand-Anything (Egonex-AI/Understand-Anything) |
78k | TS | Interactive knowledge graphs that "teach" — explore and search any codebase. Good for onboarding to unfamiliar code. |
| 5 | Codebase Memory MCP (DeusData/codebase-memory-mcp) |
— | Python | MCP server for codebase memory/context. Lighter alternative. |
| 6 | Repomix / Serena | — | — | Bundle codebases into single XML for one-off context dumps. No interactive graph. Use for quick context exports. |
Verdict: CodeGraph (installed) remains #1 for this setup. GitNexus or Graphify worth evaluating if deeper semantic structure is needed — both are MCP/skill-compatible.
2. Memory & Second Brain
Sorted by integration quality with Hermes + local-first fit.
| # | Tool | Stars | Lang | Why / Fit |
|---|---|---|---|---|
| 1 | Open Second Brain (itechmeat/open-second-brain) — INSTALLED |
~1k | JS/Bun | Native Hermes memory provider. Plain-markdown Brain/ layer, active.md injection, brain_* tools, dream pass. The best local-first second brain. Already the active provider. |
| 2 | LLM Wiki (Karpathy pattern, ~/wiki) — INSTALLED |
— | — | Interlinked markdown KB the agent reads/writes. Compounding knowledge. The user's preferred pattern (ronancodes.github.io/llm-wiki style). |
| 3 | Mem0 (mem0ai/mem0) — INSTALLED |
~40k | Python | Semantic memory with qdrant local. Configured (mem0.json), tools usable. Auto fact-extraction. |
| 4 | claude-mem (thedotmack/claude-mem) |
90k | JS | Captures everything the agent does, compresses with AI, injects relevant context across sessions. Persistent context for ANY agent (MCP-compatible). Very strong fit. |
| 5 | Supermemory (supermemoryai/supermemory) |
29k | TS | Memory + context engine, fully local option. Hermes has a supermemory provider plugin built-in (requires API key). |
| 6 | Mnemosyne (mnemosyne-oss/mnemosyne) |
2.3k | Python | Zero-cloud, SQLite-backed AI memory, one pure-Python dependency. Lightweight alternative. Hermes has a mnemosyne plugin. |
| 7 | Letta (letta-ai/letta) |
24k | Python | Platform for stateful agents with advanced memory. Heavyweight (server). |
| 8 | Graphiti (getzep/graphiti) |
30k | Python | Real-time knowledge graphs for agents. Temporal-aware. Strong for relationship memory. |
| 9 | Cognee (topoteretes/cognee) |
30k | Python | Open-source AI memory platform, persistent long-term memory. |
| 10 | agentmemory (rohitg00/agentmemory) |
27k | TS | #1 persistent memory for coding agents (benchmark-based). MCP-compatible. |
| 11 | Memory Hive (TJCurnutte/memory-hive) |
— | — | Hermes-adjacent memory plugin. |
| 12 | Brainstack (yepyhun/Brainstack) |
— | — | Second-brain stack. |
| 13 | TencentDB Agent Memory (TencentCloud/TencentDB-Agent-Memory) |
— | — | 4-tier layered memory (Working/Episodic/Semantic/Procedural). Powerful but Docker-heavy — overkill for solo dev. |
Verdict: Current stack (open-second-brain active + llm-wiki + mem0/qdrant) covers the space well. claude-mem is the standout addition candidate — it gives persistent cross-session context capture for any agent via MCP.
3. Token Optimization & Context Management
Sorted by actual value for Hermes v0.20.0.
| # | Tool | Stars | Lang | Why / Fit |
|---|---|---|---|---|
| 1 | Hermes Native Compression — BUILT-IN | — | — | compression.enabled: true (default). Summarizes older context. Use /context + /compress. The real "headroom". |
| 2 | rtk-hermes (ogallotti/rtk-hermes) — INSTALLED |
~500 | Python | Rewrites terminal commands through RTK to cut context tokens. Entry-point plugin. RTK CLI (rtk-ai/rtk) installed. |
| 3 | headroom (headroomlabs-ai/headroom) |
66k | Python | Compress tool outputs/logs/files/RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% for JSON. |
| 4 | caveman (JuliusBrussee/caveman) |
97k | JS | "Why use many token when few token do trick" — Claude Code skill cutting ~65% tokens by terse style. Not a Hermes plugin, but installable as a skill. |
| 5 | ponytail (DietrichGebert/ponytail) |
99k | JS | "Think like the laziest senior dev" — avoids writing unnecessary code. Claude Code skill. |
| 6 | hermes-tool-slimmer (alias8818/hermes-tool-slimmer) |
31 | Python | Reduces Hermes tool-schema overhead with keyword selection + Tool Search. Hermes-specific. |
| 7 | hermes-lcm (stephenschoettler/hermes-lcm) |
975 | Python | Lossless Context Management for Hermes — DAG-based context engine that never loses a message. Hermes-specific, strong fit. |
| 8 | context-mode (mksglu/context-mode) |
— | — | Context management mode. |
Verdict: Hermes compression + rtk-hermes cover the basics. hermes-lcm (975★, Hermes-native, lossless DAG context) and headroom (66k★, pre-LLM compression) are the two highest-value additions.
4. Agent Frameworks & Skills (complementary, not replacements)
Sorted by fit for extending Hermes with reusable skills.
| # | Tool | Stars | Lang | Why / Fit |
|---|---|---|---|---|
| 1 | Superpowers (obra/superpowers) |
270k | Shell | Agentic skills framework + software dev methodology. Skills system compatible with Hermes skill loading. Massive community. |
| 2 | Matt Pocock's Skills (mattpocock/skills) |
211k | Shell | "Skills for Real Engineers" — production-grade engineering skills. |
| 3 | Addy Osmani's Agent Skills (addyosmani/agent-skills) |
85k | JS | Production-grade engineering skills for AI coding agents. |
| 4 | Anthropic Skills (anthropics/skills) |
— | — | Official skills collection. |
| 5 | Karpathy Skills (multica-ai/andrej-karpathy-skills) |
— | — | Skills distilled from Karpathy's approach. |
| 6 | OpenSpec (Fission-AI/OpenSpec) |
— | — | Spec-driven development. |
| 7 | Agent Skills Hub (multica-ai/multica) |
— | — | Skills aggregator. |
Verdict: Hermes already has a rich skills system. Superpowers is the standout — its methodology + skills complement Hermes's own skills nicely.
5. Agent Runtimes & Alternatives (context)
These are NOT needed for this setup but appear in the ecosystem graph.
| Tool | Stars | Why it's here |
|---|---|---|
OpenClaw (openclaw/openclaw) |
386k | Popular agent framework; Hermes's related repo. |
OpenCode (anomalyco/opencode) |
195k | Open-source coding agent. |
pi (earendil-works/pi) |
86k | AI agent toolkit (unified LLM API, TUI, coding CLI). |
Claude Code (anthropics/claude-code) |
— | Anthropic's coding agent. |
OpenAI Codex (openai/codex) |
— | OpenAI's coding agent. |
6. RAG / Retrieval (for future semantic features)
If FicHub's ask-the-archive / semantic search grows:
| Tool | Stars | Lang | Why / Fit |
|---|---|---|---|
RAGFlow (infiniflow/ragflow) |
87k | Go | Leading open-source RAG engine. Heavy (server). |
Microsoft GraphRAG (microsoft/graphrag) |
35k | Python | Graph-based RAG. Strong for relationship-heavy retrieval. |
Crawl4AI (unclecode/crawl4ai) |
78k | Python | LLM-friendly web crawler/scraper — great for FicHub ingestion. |
Firecrawl (firecrawl/firecrawl) |
164k | TS | The "context API" for search/scrape/interact at scale. |
7. MCP Ecosystem (the protocol layer)
| Tool | Stars | Why |
|---|---|---|
MCP Servers (modelcontextprotocol/servers) |
89k | Official reference servers. |
Awesome MCP Servers (punkpeye/awesome-mcp-servers) |
92k | The definitive catalog — browse for any integration. |
browser-use (browser-use/browser-use) |
— | Browser automation (Hermes has a browser-use plugin bundled). |
8. Already-Installed Summary (this box)
| Component | Type | Status |
|---|---|---|
| CodeGraph | MCP server | Installed, FicHub indexed |
| open-second-brain | Memory provider | Active |
llm-wiki (~/wiki) |
Skill + KB | Active |
| mem0 + qdrant | Memory provider (semantic) | Configured |
| rtk-hermes + RTK CLI | Token plugin | Enabled |
| LSP (rust-analyzer, svelte, ts, pyright) | Built-in subsystem | Installed |
| Ollama (llama3.1:8b, nomic-embed-text) | Local LLM/embedder | Running |
Recommended additions (highest value first)
- claude-mem — persistent cross-session context capture via MCP. Complements all three memory layers.
- hermes-lcm — lossless DAG context management, Hermes-native.
- headroom — pre-LLM output compression (66k★, MCP-compatible).
- Superpowers skills — methodology + skills that slot into Hermes's skill loader.
- GitNexus or Graphify — if CodeGraph's tree-sitter approximation ever feels limiting.
Not worth it for this setup
- TencentDB Agent Memory — Docker-heavy, overkill solo.
- Letta / Cognee / RAGFlow servers — heavyweight infra; mem0/qdrant + llm-wiki already cover memory.