# Hermes Agent Namespace Instructions This is a compiled namespace source under `pixi-vault/wikis/hermes-agent/`. ## Rules - Follow the root `Wiki Compiler Maps/Namespace Wiki Compiler Map.md`. - Keep this namespace scoped to Hermes Agent usage, setup, configuration, MCP, traces, evals, observability, and Jamie/Pixoid operating reference. - Source priority is explicit: official Hermes docs and local installed Hermes source outrank AgentWikis. AgentWikis is a curated discovery map, not source-of-truth for commands. - Do not import the whole AgentWikis Hermes wiki. Curate only pages Jamie's workflow uses. - Update `wiki/index.md` and `wiki/log.md` whenever compiled pages are added. --- title: Hermes Agent created: 2026-06-18 updated: 2026-06-27 type: namespace-overview status: compiled category: agent-infrastructure namespace: hermes-agent confidence: medium --- # Hermes Agent > Curated Hermes Agent operating reference for Jamie/Pixoid: setup, configuration, MCP, profiles, tracing, evals, observability, and workflow-specific usage. ## Scope ### Covers Hermes Agent setup and configuration, MCP integration, profiles, skills, memory, cron, gateway operations, subagents/delegation, session traces, trajectory capture, batch evals, Langfuse traces, NeMo Relay ATOF/ATIF exports, capability-routing decisions, and Jamie-specific operating reference for Pixoid. ### Not Covered General AI-agent theory, unrelated community projects, old video transcripts unless they explain a workflow Jamie uses, stale release history not needed for current operation, and web rumors not verified against official Hermes docs or local installed source. ### Current As 2026-06-27 — added Hermes Remote Artifact Previews as a reusable rule: remote Desktop artifacts must use dashboard HTTP/API, never VPS-local `file://` preview paths. ## Source Priority 1. Official Hermes docs: `https://hermes-agent.nousresearch.com/docs` 2. Local installed Hermes source: `/usr/local/lib/hermes-agent` 3. AgentWikis Hermes: `https://agentwikis.com/wiki/hermes/llms.txt` 4. Web search for freshness checks only ## Canonical Source Roots - Official docs mirrored in local install: `/usr/local/lib/hermes-agent/website/docs/` - Canonical capability-routing concept: `Knowledge/concepts/hermes-capability-routing.md` - Remote artifact preview rule: `Knowledge/concepts/hermes-remote-artifact-previews.md` - Local source/plugin docs: `/usr/local/lib/hermes-agent/` - AgentWikis Hermes raw pages: `https://agentwikis.com/raw/hermes/...` ## Crosslinks - [[../agent-workflows/README|agent-workflows]] - [[../eval-trace/README|eval-trace]] - [[../pixi-vault/README|pixi-vault]] - [[../local-ai-infrastructure/README|local-ai-infrastructure]] ## Maintenance - Verify commands against official docs or live `hermes --help` before using them. - Prefer MCP retrieval over web search for stable Hermes reference questions. - Use web search only when checking latest releases, GitHub issues, or current external state. - Keep this namespace slim: add operational pages, not a full AgentWikis mirror. --- title: Hermes Evals and Traces Setup created: 2026-06-18 updated: 2026-06-18 type: concept status: compiled namespace: hermes-agent tags: [hermes-agent, evals, traces, observability, langfuse, nemo-relay, trajectories] sources: - https://hermes-agent.nousresearch.com/docs/user-guide/features/batch-processing - https://hermes-agent.nousresearch.com/docs/user-guide/features/built-in-plugins - /usr/local/lib/hermes-agent/website/docs/developer-guide/trajectory-format.md - /usr/local/lib/hermes-agent/plugins/observability/nemo_relay/README.md confidence: high --- # Hermes Evals and Traces Setup Hermes has native trace surfaces and trajectory capture, but it does not ship a full generic pass/fail eval dashboard. Use the right layer for the job. ## Native trace layers | Layer | Use | Setup | |---|---|---| | `state.db` sessions | Review real conversations, tool calls, usage, cost, and platform history. | Built in. Use `hermes sessions ...`. | | Session JSONL export | Build eval datasets or review failure traces outside Hermes. | `hermes sessions export ...`. | | `/insights` | Usage analytics: models, tools, skills, tokens, cost, platforms, top sessions. | Built in. Use `hermes insights --days N`. | | Trajectory saving | ShareGPT-compatible turn/tool traces from CLI sessions. | `agent.save_trajectories: true` or `--save-trajectories`. | | Batch runner | Run many prompts and collect structured trajectories/statistics. | `python batch_runner.py ...`. | | Langfuse plugin | External LLM observability dashboard for turns, LLM calls, and tool calls. | Enable `observability/langfuse` and set keys. | | NeMo Relay plugin | ATOF/ATIF exports for replay, evaluation, and harness analysis. | Enable `observability/nemo_relay` and set export env vars. | ## Session export ```bash hermes sessions export sessions.jsonl hermes sessions export --source discord discord.jsonl hermes sessions export --session-id one-session.jsonl ``` Use this for failure review, dataset building, replay-style analysis, and tool-use trace mining. ## Insights ```bash hermes insights --days 30 hermes insights --days 7 --source cli ``` Use this for operational analytics, not task correctness scoring. ## Trajectory saving Official developer docs describe ShareGPT-compatible JSONL trajectories written as: ```text trajectory_samples.jsonl # completed=True failed_trajectories.jsonl # failed or interrupted ``` Enable in config: ```yaml agent: save_trajectories: true ``` or run with: ```bash hermes chat --save-trajectories -q "..." ``` The batch runner always saves trajectories. ## Batch eval runner Official batch-processing docs describe `batch_runner.py` as the native path for running many prompts and producing trajectory data plus statistics. ```bash python batch_runner.py --dataset_file=data/eval_suite.jsonl --batch_size=10 --run_name=eval_gpt55 --model=gpt-5.5 --num_workers=4 --max_turns=10 ``` Outputs: ```text data// ├── trajectories.jsonl ├── batch_0.jsonl ├── checkpoint.json └── statistics.json ``` Tracked fields include `completed`, `partial`, `api_calls`, `toolsets_used`, `tool_stats`, and `tool_error_counts`. ## Langfuse setup Official built-in plugin docs say the Langfuse plugin traces Hermes turns, LLM calls, and tool invocations. Interactive setup: ```bash hermes tools # choose Langfuse Observability ``` Manual setup: ```bash pip install langfuse hermes plugins enable observability/langfuse ``` Add credentials to `~/.hermes/.env`: ```bash HERMES_LANGFUSE_PUBLIC_KEY=pk-lf-... HERMES_LANGFUSE_SECRET_KEY=sk-lf-... HERMES_LANGFUSE_BASE_URL=https://cloud.langfuse.com ``` Verify: ```bash hermes plugins list hermes chat -q "hello" ``` Then check Langfuse for a `Hermes turn` trace. ## NeMo Relay ATOF/ATIF setup The NeMo Relay plugin maps Hermes observer hooks to scopes, LLM spans, tool spans, marks, ATOF, and ATIF. It can export raw lifecycle events as ATOF JSONL and ATIF trajectory files for replay/eval harnesses. Enable plugin: ```bash hermes plugins enable observability/nemo_relay ``` Install runtime if needed: ```bash pip install "nemo-relay==0.3" ``` Local export settings: ```bash export HERMES_NEMO_RELAY_ATOF_ENABLED=1 export HERMES_NEMO_RELAY_ATOF_OUTPUT_DIRECTORY=.nemo-relay/atof export HERMES_NEMO_RELAY_ATIF_ENABLED=1 export HERMES_NEMO_RELAY_ATIF_OUTPUT_DIRECTORY=.nemo-relay/atif ``` Optional controls include `HERMES_NEMO_RELAY_ATOF_FILENAME`, `HERMES_NEMO_RELAY_ATOF_MODE`, `HERMES_NEMO_RELAY_ATIF_FILENAME_TEMPLATE`, `HERMES_NEMO_RELAY_ATIF_AGENT_NAME`, `HERMES_NEMO_RELAY_ATIF_AGENT_VERSION`, `HERMES_NEMO_RELAY_ATIF_MODEL_NAME`, and `HERMES_NEMO_RELAY_ATIF_SUBAGENT_EXPORT_MODE`. ## Decision rule Use the lightest trace surface that answers the question: ```text Need to inspect one past run? → sessions export / session_search Need usage trends? → insights Need many-prompt eval trajectories? → batch_runner.py Need external observability UI? → Langfuse Need replay/harness trajectory form?→ NeMo Relay ATIF/ATOF ``` --- title: Hermes Capability Routing created: 2026-06-26 updated: 2026-06-26 type: concept status: compiled namespace: hermes-agent source: Knowledge/concepts/hermes-capability-routing.md confidence: high --- # Hermes Capability Routing Hermes capability routing is the decision lens for choosing the smallest effective Hermes Agent surface for a task: direct tool call, `execute_code`, skill, memory, session search, delegation, cron, gateway, API server, MCP, plugin, profile, kanban board, or external process. Short form: ```text intent → smallest surface → durability boundary → side-effect boundary → verification handle ``` ## Why this exists Hermes exposes many overlapping surfaces. The optimal route depends on task shape, durability, side effects, and verification burden, not on which feature sounds most powerful. Use this when a request is broader than skill selection but narrower than a full route contract. Skill routing decides which procedures frame the work. Capability routing decides whether the work should be a tool call, script, skill, subagent, scheduled job, gateway/API integration, profile route, MCP/plugin extension, or knowledge/project update. ## Routing matrix | Task shape | Default Hermes surface | |---|---| | One immediate read/write/check | Current-session tools | | Mechanical multi-tool loop | `execute_code` with `hermes_tools` | | Reusable procedure | Hermes skill | | Stable user/environment fact | Memory/user profile | | Past-session recall | `session_search` | | Independent bounded analysis | `delegate_task` | | Durable scheduled work | `cronjob` or tracked background process | | Repeated multi-profile collaboration | Kanban/profile route | | Messaging delivery | Gateway/platform adapter | | External app needs OpenAI-compatible access | API server | | Existing external service has a server | MCP | | New reusable action | Tool or plugin | | Cost/speed/reliability choice | Model picker, provider routing, fallback providers | ## Public wiki placement Primary namespace: `hermes-agent`, because this is a Hermes Agent capability-selection map derived from official docs and the reusable local skill. Relevant crosslink namespace: `agent-workflows`, because Pixoid crew routing uses the lens to decide when work should become skill work, subagent work, scheduled work, peer-profile/kanban work, vault/Pixi Wiki source work, or a public deploy candidate. Do not split this into a new namespace; it connects existing Hermes Agent and Agent Workflows pages. ## Boundaries - Official Hermes docs remain source of truth for commands, flags, providers, config keys, and setup details. - Capability routing is not authorization; destructive commands, deploys, merges, secrets, profile/gateway/provider/webhook changes, and public publication still require the relevant approval boundary. - `delegate_task` is not durable; use cron/background/kanban/profile routes for work that must survive interruption. - Do not make memory carry procedures, project state, or public wiki content. - Do not build custom plugins/tools until existing CLI, tools, skills, MCP, gateway, and API surfaces have been checked. ## Related pages - [[../../agent-workflows/wiki/concepts/agent-skill-routing|Agent Skill Routing]] - [[../../agent-workflows/wiki/concepts/agent-capability-route-pattern|Agent Capability Route Pattern]] - [[../../agent-workflows/wiki/concepts/runtime-memory-knowledge-routing|Runtime Memory Knowledge Routing]] - [[../../agent-workflows/wiki/concepts/profile-memory-boundaries|Profile Memory Boundaries]] - [[concepts/source-priority|Hermes Source Priority]] ## Sources - Official docs: `https://hermes-agent.nousresearch.com/docs` - Canonical source: `Knowledge/concepts/hermes-capability-routing.md` - Reusable skill: `~/.hermes/skills/autonomous-ai-agents/hermes-capability-routing/SKILL.md` --- title: Hermes Remote Artifact Previews created: 2026-06-27 updated: 2026-06-27 type: concept status: compiled namespace: hermes-agent source: Knowledge/concepts/hermes-remote-artifact-previews.md confidence: high --- # Hermes Remote Artifact Previews Hermes remote artifact previews are HTML/MDX/file artifacts generated on a remote Hermes host and opened from a local Hermes Desktop or dashboard session. ## Core rule Never hand a remote Desktop session a VPS-local `file://` URL such as: ```text file:///tmp/.../artifact.html ``` Use the dashboard HTTP/API boundary instead: ```text remote file on VPS -> dashboard HTTP/API -> local Desktop preview ``` For Jamie's VPS setup: - dashboard origin: `http://100.82.175.2:9119` - gateway/API port: `8642` remains separate - artifact links should be verified `http://...` URLs, not `file://...` paths ## Why this matters A local Electron/Chromium webview cannot read the VPS filesystem. A dashboard page loaded over HTTP also cannot load arbitrary local resources through `file://`; Chromium reports this as `Not allowed to load local resource` or `chromewebdata`. ## Durable implementation pattern For product behavior, remote HTML artifacts should be read or served through the Hermes dashboard backend: - read remote HTML through the authenticated dashboard filesystem API and render a safe data/HTTP preview target; or - serve the artifact through a dashboard/static route and return that HTTP URL. Temporary `python3 -m http.server` links are acceptable for demos, but they are not durable product infrastructure. ## Agent operating checklist 1. Inspect topology: local Desktop vs remote VPS. 2. Treat `/tmp/...`, `/root/...`, and similar paths as remote-only when the Desktop is local. 3. Do not paste `file://` as the user-facing preview link. 4. Produce a dashboard HTTP/API URL or explicitly start a temporary HTTP endpoint. 5. Verify with `curl -I` that the URL is `200 OK` before saying it works. ## Related pages - [[concepts/hermes-capability-routing|Hermes Capability Routing]] - [[concepts/source-priority|Hermes Source Priority]] - [[../../agent-workflows/wiki/concepts/runtime-memory-knowledge-routing|Runtime Memory Knowledge Routing]] --- title: Hermes Session Trace Store created: 2026-06-18 updated: 2026-06-18 type: concept status: compiled namespace: hermes-agent tags: [hermes-agent, sessions, traces, state-db, insights] sources: - /usr/local/lib/hermes-agent/hermes_state.py - /usr/local/lib/hermes-agent/agent/insights.py - https://hermes-agent.nousresearch.com/docs/reference/cli-commands confidence: high --- # Hermes Session Trace Store Hermes' built-in trace substrate is the profile-scoped SQLite session store. ## What it stores `state.db` stores session metadata, full message history, model configuration, tool calls, token usage, costs, source/platform tags, and searchable message text via FTS5. The local source describes it as replacing older per-session JSONL storage for normal CLI/gateway sessions. Batch runner and RL trajectories are separate systems. ## Useful commands ```bash hermes sessions list hermes sessions browse hermes sessions stats hermes sessions export out.jsonl hermes sessions export --source discord discord.jsonl hermes sessions export --session-id one-session.jsonl ``` ## Insights `hermes insights` analyzes `state.db` for token consumption, cost estimates, tool usage patterns, skill usage, activity trends, model/platform breakdowns, and top sessions. ```bash hermes insights --days 30 hermes insights --days 7 --source cli ``` ## Boundary This is operational telemetry. It is not a correctness judge. Use it to find traces, mine examples, and inspect behavior; pair it with explicit eval prompts or human review when measuring quality. --- title: Hermes Source Priority created: 2026-06-18 updated: 2026-06-18 type: concept status: compiled namespace: hermes-agent tags: [hermes-agent, source-priority, docs, agentwikis] sources: - https://hermes-agent.nousresearch.com/docs - /usr/local/lib/hermes-agent - https://agentwikis.com/wiki/hermes/llms.txt confidence: high --- # Hermes Source Priority Hermes reference work should separate **truth** from **maps**. ## Priority order 1. **Official Hermes docs** — commands, setup, configuration, current documented behavior. 2. **Local installed Hermes source** — what this VPS actually has installed and can run. 3. **AgentWikis Hermes** — curated topic map, release summaries, ecosystem inventory, and explanation layer. 4. **Web search** — freshness checks for new releases, open issues, and current community state. ## AgentWikis role AgentWikis is useful as a discovery map. It surfaces topics such as MCP integration, configuration, profiles, subagents, release summaries, providers, memory backends, and community projects. It should not override official docs or live source when commands differ. ## Use policy Use AgentWikis to decide what to read next. Use official docs or live source to decide what to run. ```text Question: "What Hermes features exist?" → AgentWikis is useful. Question: "What command should I run now?" → Official docs/live CLI first. Question: "Why is this VPS failing?" → Inspect live state first. Question: "Is this still current upstream?" → Official docs + web/GitHub check. ``` ## Import boundary Do not mirror the whole AgentWikis Hermes corpus into Pixi Wiki. Import only pages that support Jamie/Pixoid operations, and preserve source metadata plus freshness dates. --- title: Hermes Agent — Master Index created: 2026-06-18 updated: 2026-06-27 type: index status: compiled namespace: hermes-agent --- # Hermes Agent — Master Index > Compiled index for `hermes-agent`. ## Concepts - [[concepts/source-priority|Hermes Source Priority]] — How to rank official docs, local source, AgentWikis, and web search. - [[concepts/hermes-capability-routing|Hermes Capability Routing]] — Select the smallest effective Hermes surface for a task: tool, script, skill, memory, delegation, cron, gateway/API, MCP/plugin, profile, kanban, provider route, and verification handle. - [[concepts/hermes-remote-artifact-previews|Hermes Remote Artifact Previews]] — Use dashboard HTTP/API for remote Desktop artifacts; never hand local Desktop a VPS `file://` path. - [[concepts/evals-and-traces-setup|Hermes Evals and Traces Setup]] — Practical setup map for session exports, batch trajectories, Langfuse, and NeMo Relay ATOF/ATIF. - [[concepts/session-trace-store|Hermes Session Trace Store]] — Native `state.db`, session export, and `/insights` surfaces. ## Entities ## Summaries - [[summaries/agentwikis-hermes-curated-seed|AgentWikis Hermes Curated Seed]] — What to use from AgentWikis and what to skip. - [[summaries/external-hermes-wikis-import-review|External Hermes Wikis Import Review]] — Triage of Hermes Guide and AgentWikis content worth importing for Jamie/Pixoid. ## Syntheses ## Source Roots - Official docs: `https://hermes-agent.nousresearch.com/docs` - Canonical concept: `Knowledge/concepts/hermes-capability-routing.md` - Local installed source: `/usr/local/lib/hermes-agent` - AgentWikis Hermes: `https://agentwikis.com/wiki/hermes/llms.txt` --- title: Hermes Agent — Activity Log created: 2026-06-18 updated: 2026-06-27 type: log status: compiled namespace: hermes-agent --- # Hermes Agent — Activity Log ## 2026-06-27 - Added `wiki/concepts/hermes-remote-artifact-previews.md` to preserve the remote Desktop preview rule: artifacts generated on the VPS must cross the dashboard HTTP/API boundary, not `file://`. - Updated namespace index/README so future agents retrieve this rule before handing Jamie artifact URLs. ## 2026-06-26 - Added `wiki/concepts/hermes-capability-routing.md` as the primary Hermes capability-selection concept from the official docs review and reusable local skill. - Updated namespace README and index to route capability questions to `hermes-agent`; `agent-workflows` carries a cross-namespace pointer for Pixoid crew routing. - Preserved the public deploy boundary: source updated here, generated public `pixi-wiki` requires verified rebuild and explicit deploy/push approval. ## 2026-06-18 - Seeded `hermes-agent` namespace as a slim operating reference, not a wholesale AgentWikis mirror. - Added source-priority policy: official docs and local installed source outrank AgentWikis; AgentWikis is a discovery map. - Added setup notes for native session traces, batch trajectories, Langfuse traces, and NeMo Relay ATOF/ATIF exports from official Hermes docs/local source. - Reviewed Hermes Guide and AgentWikis Hermes for import candidates; added `wiki/summaries/external-hermes-wikis-import-review.md` with a slim import queue and skip rules. ## 2026-06-19 - Re-ran the external Hermes wiki import assessment after namespace population updates. - Confirmed no wholesale import and no new immediate import pages; updated `wiki/summaries/external-hermes-wikis-import-review.md` with current namespace routing, crosslink-only/default defer decisions, and source-priority caveats. --- title: AgentWikis Hermes Curated Seed created: 2026-06-18 updated: 2026-06-18 type: summary status: compiled namespace: hermes-agent tags: [hermes-agent, agentwikis, source-ingest, curation] sources: - https://agentwikis.com/wiki/hermes/README.md - https://agentwikis.com/wiki/hermes/llms.txt - https://agentwikis.com/raw/hermes/wiki/index.md confidence: medium --- # AgentWikis Hermes Curated Seed AgentWikis hosts a Hermes knowledge base with roughly 105 documents covering Hermes setup, configuration, workflows, extensions, releases, memory providers, platforms, providers, and community projects. ## Useful slices for Jamie/Pixoid Use AgentWikis as a map for: - MCP integration - CLI and configuration reference - profiles and multi-instance routing - skills and memory systems - cron and scheduling - subagents and delegation - release summaries when checking feature history - ecosystem/community inventory when exploring external references ## Skip by default Do not import everything. Skip or quarantine: - video transcript summaries unless a transcript explains a workflow Jamie uses; - broad community lists unless they answer a current question; - old release pages unless debugging historical behavior; - pages that conflict with official docs or local installed source. ## Staleness note The AgentWikis Hermes README still claims an older verified release in one place, while `llms.txt` and the master index report v0.16.0 current as of 2026-06-16. Treat this as a useful but secondary source. ## Import posture AgentWikis is a discovery layer. Official Hermes docs and live local source decide commands, config, and setup steps. --- title: External Hermes Wikis Import Review created: 2026-06-18 updated: 2026-06-19 type: summary status: compiled namespace: hermes-agent tags: [hermes-agent, curation, import-review, agentwikis, hermesguide] sources: - https://hermesguide.xyz/wiki/ - https://hermesguide.xyz/directory - https://agentwikis.com/wiki/hermes/README.md - https://agentwikis.com/wiki/hermes/llms.txt - https://agentwikis.com/raw/hermes/wiki/index.md confidence: medium --- # External Hermes Wikis Import Review Reviewed `hermesguide.xyz` and AgentWikis Hermes for content worth importing into Jamie/Pixoid's Hermes Agent namespace. ## Verdict Do **not** mirror either source. Keep this namespace slim and operational. Import only content that is: - hard to rediscover with one web search; - likely to help Jamie operate Hermes, profiles, gateway, cron, memory, evals, or subagents; - stable enough to be useful after a week; - cross-checked against official Hermes docs or local installed source before commands/config are trusted. ## Hermes Guide / hermesguide.xyz ### Useful signal `https://hermesguide.xyz/wiki/` is currently a landing/marketing route, not a deep wiki. The useful page is the ecosystem directory at `https://hermesguide.xyz/directory`, which lists 156+ community tools grouped by Dev Workflow, Integrations, Personal Assistant, Meta & Ecosystem, Business Ops, Enterprise, Content Creation, Cost Optimization, Creative, Research, Messaging, Trading & Markets, and Privacy/Self-Hosted. ### Worth importing Import as **one summary**, not 156 entity pages: - `Hermes Ecosystem Directory Triage` — curated shortlist of tools Jamie might actually evaluate. - Include categories, short descriptions, and triage status: `watch`, `maybe`, `skip`, `needs verification`. - Verify each selected tool against its primary repo before creating an entity page. Possible shortlist for Jamie: - Hermes Desktop / Scarf / Hermes Workspace / Hermes Console — front-end/control-plane options for sessions, memory, cron, project dashboards, and SSH-first operation. - Hermes Memory UI / Herm / Autograph / Obsidian memory keep-alive — memory observability or Obsidian-linked memory workflows. - Hermes Labyrinth / Hermescheck / Hermes progress tail / H-Ops — observability, health checks, progress visibility, and Kanban operations. - Hermes Google Workspace / Microsoft Graph API skill / Nextcloud / Apple Calendar assistant — productivity integrations matching Jamie's assistant workflows. - Hermes-Web-Search-Plus / video research ingest / Obsidian-Video-Notes — research ingestion helpers. - Hermes-Multitenancy / Shellward / Hermes-Aegis / HermesClaw — security and multi-tenant routing ideas, only if Jamie works on gateway/profile isolation. - Agent-Team-Orchestrator / Maestro / mission-control-like tools — compare against Jamie's existing crew/mission-control model; avoid importing unless they add a concrete pattern. ### Skip - Broad creative/game/trading novelty tools unless Jamie opens a matching project. - Most named forks with no current need. - Pricing/model benchmark pages from Hermes Guide; freshness decays fast and web search is better. - Landing pages that duplicate official docs. ## AgentWikis Hermes AgentWikis has roughly 105 Hermes documents. It is a useful discovery map, but it is secondary to official docs and local source. ### Already covered locally The current `hermes-agent` namespace already has: - [[../concepts/source-priority|Hermes Source Priority]] - [[../concepts/session-trace-store|Hermes Session Trace Store]] - [[../concepts/evals-and-traces-setup|Hermes Evals and Traces Setup]] - [[agentwikis-hermes-curated-seed|AgentWikis Hermes Curated Seed]] ### Highest-value imports Create or expand these only when verified against official docs/local source: 1. `Memory Providers Compared` — high leverage because provider choice is hard to rediscover by search and crosses Honcho, Holographic, Hindsight, Mem0, OpenViking, ByteRover, RetainDB, Supermemory, and Memori. This should be a concept or synthesis, not separate entity pages for every provider unless Jamie evaluates them. 2. `Deployment Backends Compared` — useful for local vs Docker vs SSH/VPS vs Modal vs Daytona vs Singularity decisions. Jamie already uses VPS/WSL/local distinctions, so this would reduce repeated setup reasoning. 3. `Cron Troubleshooting Checklist` — worth a compact operational concept: timing, delivery, permissions, skill loading. This is likely to be reused for crew cron and gateway jobs. 4. `Auxiliary Models` — import the footguns: side-task routing, `auto` fallback, compression config split, approval/vision/web_extract/curator/task slots. This is web-search-resistant operational detail. 5. `Profiles & Multi-Instance` — import as Jamie-specific operating notes: profile isolation, gateway token boundaries, profile-scoped cron/memory paths, when to clone vs separate profile. 6. `Subagents & Delegation` — import only the operational decision table: `delegate_task` vs background process vs cron vs full spawned Hermes instance. 7. `Local Stack / Airplane Mode` — import only if Jamie actively works on local/offline Hermes; otherwise keep as a deferred pointer under local-ai-infrastructure. ### Medium priority / import on demand - `Skills Catalog — 644-Skill Map` — useful as a pointer, but likely stale; prefer live `hermes skills browse/search` when available. - `Awesome Hermes Agent — Ecosystem Inventory` — overlaps Hermes Guide directory; keep as source for the same ecosystem triage summary. - Release summaries — only import when debugging historical behavior. - Provider pages — import only selected providers Jamie actually uses or evaluates. - Platform pages — import only Discord/Telegram/Email/Home Assistant/Google Workspace/Slack if actively configured. ### Skip by default - Video transcript summaries unless they contain a workflow Jamie wants to preserve. - OpenClaw comparison pages unless doing migration/positioning work. - Onchain/trading workflow pages. - One-page entity stubs for every community project. - Old version pages as standalone entities. ## Suggested import queue 1. Create `wiki/syntheses/memory-providers-compared.md` from AgentWikis + official docs. Crosslink to agent-workflows memory-routing pages. 2. Create `wiki/syntheses/hermes-deployment-backends-compared.md` from AgentWikis + local source/docs. Include Jamie-specific VPS/WSL/Puffer notes only if they are stable and already authorized for this namespace. 3. Create `wiki/concepts/hermes-cron-troubleshooting.md` as a compact four-category checklist. 4. Create `wiki/concepts/hermes-auxiliary-model-routing.md` for side-task model routing and compression footguns. 5. Create `wiki/summaries/hermes-ecosystem-directory-triage.md` from Hermes Guide + AgentWikis ecosystem pages, with selected tools only. ## Import rule One external source can justify a local page only if the page answers: “Would Jamie/Pixoid plausibly ask this again, and would web search be slower or noisier than retrieving the wiki page?” If yes, import a compact synthesis. If no, leave the source as a pointer. ## Namespace routing review Primary home stays `hermes-agent` for Hermes-specific operations and setup. Other namespaces should crosslink to this review or to future focused pages rather than duplicate the content. | Candidate | Primary namespace | Crosslink namespaces | Decision | |---|---|---|---| | Memory Providers Compared | `hermes-agent` | `agent-workflows`, `local-ai-infrastructure` | Add as a Hermes synthesis if Jamie starts comparing providers; crosslink to runtime memory routing and local-first memory infrastructure. | | Deployment Backends Compared | `hermes-agent` | `local-ai-infrastructure`, `agent-workflows` | Add as a Hermes synthesis; crosslink from local infrastructure because backend choice affects VPS/WSL/local execution. | | Cron Troubleshooting Checklist | `hermes-agent` | `agent-workflows` | Add as a Hermes operational concept; crosslink from crew/cron workflow pages because recurring jobs are workflow infrastructure. | | Auxiliary Models | `hermes-agent` | `eval-trace`, `agent-workflows` | Add as a Hermes concept; crosslink where eval/tracing side tasks and approval/compression behavior affect workflow quality. | | Profiles & Multi-Instance | `hermes-agent` | `agent-workflows` | Add as a Hermes concept; crosslink from profile/memory boundary and peer-profile pages. | | Subagents & Delegation | `hermes-agent` | `agent-workflows` | Split if needed: Hermes API/tool mechanics in `hermes-agent`; route-governance decision table in `agent-workflows`. | | Local Stack / Airplane Mode | `local-ai-infrastructure` | `hermes-agent` | If imported, primary home should be local infrastructure; Hermes Agent should only link to setup-specific notes. | | Ecosystem Directory Triage | `hermes-agent` | `pixi-vault`, `agent-workflows`, `local-ai-infrastructure` | Keep as a single triage summary first; create entity pages only for tools Jamie evaluates. | | AgentWikis / external wiki import pattern | `pixi-vault` | `hermes-agent`, `agent-workflows` | The reusable import/routing method belongs to Pixi Vault; this page is the Hermes-specific instance. | ### Crosslink policy - `hermes-agent` owns Hermes commands, config, providers, memory plugins, profiles, gateway, cron, skills, and subagent mechanics. - `agent-workflows` owns Pixoid/Tinker/Quill/Boba route decisions, handoffs, governance, memory-boundary behavior, and when to delegate vs spawn vs schedule. - `local-ai-infrastructure` owns offline/local model stacks, local retrieval, hardware constraints, and local-first deployment tradeoffs. - `eval-trace` owns observability/evaluation failure modes and evidence gates, not general Hermes setup. - `pixi-vault` owns the namespace-routing/import methodology and public wiki/compiler mechanics. --- ## 2026-06-19 re-run after namespace population updates ### Live Pixi Wiki state that changes routing The namespace set is now materially populated, so the external Hermes sources should be routed more narrowly than the first review: - `hermes-agent` already contains source-priority, session-trace, eval/trace setup, AgentWikis seed, and this import review. It remains the primary home for Hermes commands, config, providers, memory, profiles, gateway, cron, skills, and subagent mechanics. - `agent-workflows` is now a real workflow namespace with route-pattern, SOUL.md, memory-boundary, peer-profile, Ponytail, and crew operating model pages. It should receive crosslinks for governance decisions, not duplicate Hermes mechanics. - `local-ai-infrastructure` now has local retrieval and RAG-over-AgentWikis pages. It is the right primary home for local/offline model stack tradeoffs, with Hermes setup notes linked back to `hermes-agent`. - `eval-trace` now has context-overfitting, workflow-quality evaluation, and Eval Trace entity pages. It should crosslink auxiliary-model/evidence-gate footguns, not own general Hermes setup. - `pixi-vault` now has Pixi Wiki and compiler-model pages. It owns the reusable import methodology and source/output repo boundary, not Hermes operations. - Newly populated `ai-native-product-surfaces`, `curated-tuning-datasets`, and `rl-sim-labs` do not materially change this import route. They are useful only if a specific Hermes tool becomes part of a product demo, dataset pipeline, or RL experiment workflow. ### Source inspection notes - Hermes Guide `/wiki/` is still mostly landing/marketing HTML. Its reusable signal remains `/directory`, now observed as a 156+ tool directory across Dev Workflow, Integrations, Personal Assistant, Meta & Ecosystem, Business Ops, Enterprise, Content Creation, Cost Optimization, Creative, Research, Messaging, Trading & Markets, and Privacy/Self-Hosted. - AgentWikis Hermes still exposes about 105 documents through `llms.txt` / raw markdown. It has useful topic maps for auxiliary models, deployment backends, cron troubleshooting, profiles, subagents, memory, local models, release notes, providers, platforms, and ecosystem pages. - Any command/config/runtime claim from either external source still requires official Hermes docs or local installed source verification before use. ### Candidate imports by priority #### Import now No new full imports recommended from the re-run. The high-reuse material is already represented as source-priority, curated seed, eval/trace setup, and this routing review. Importing more now would mostly duplicate official docs or AgentWikis pages without a live operating need. #### Crosslink only - `Subagents & Delegation` — crosslink from `agent-workflows` route-governance pages to the Hermes mechanics in `hermes-agent`; do not duplicate the API mechanics in workflow pages. - `Profiles & Multi-Instance` — crosslink from `profile-memory-boundaries`, `peer-profiles-vs-child-processes`, and `hermes-soul-md-wiring` to a future Hermes mechanics page if Jamie asks for one. - `Local Models & Airplane Mode` — crosslink from `local-ai-infrastructure`; import only if local/offline Hermes operation becomes an active slice. - `Auxiliary Models` — crosslink from `eval-trace` for evidence gates and context-overfit diagnostics; import only a Hermes concept if troubleshooting side-task routing. - `Hermes Guide Directory` and AgentWikis ecosystem pages — keep as a source pointer for future triage; do not create 156 entity pages. #### Import on demand - `Memory Providers Compared` — synthesize only when choosing or troubleshooting Honcho/Mem0/Holographic/Hindsight/etc.; primary `hermes-agent`, crosslink `agent-workflows` and `local-ai-infrastructure`. - `Deployment Backends Compared` — synthesize only when making a VPS/WSL/local/Docker/Modal/Daytona choice; primary `hermes-agent`, crosslink `local-ai-infrastructure` and `agent-workflows`. - `Hermes Cron Troubleshooting` — create only when a real cron job fails or crew cron behavior needs a durable checklist; primary `hermes-agent`, crosslink `agent-workflows`. - `Hermes Auxiliary Model Routing` — create only when side-task routing/compression/vision/web-extract behavior affects work quality; primary `hermes-agent`, crosslink `eval-trace`. - `Hermes Ecosystem Directory Triage` — create only when Jamie asks to evaluate external Hermes tools; primary `hermes-agent`, crosslink `pixi-vault` for import method and `agent-workflows` for route/governance patterns. #### Skip - Wholesale AgentWikis mirror. - Hermes Guide landing/marketing pages. - Video transcripts unless a specific workflow becomes durable. - Release/version pages unless debugging historical behavior. - Provider/platform pages unless Jamie is configuring that provider/platform. - Onchain/trading/game/novelty ecosystem tools unless a matching project is opened. - Entity pages for every memory provider, backend, community project, or directory listing. ### Updated routing table | Candidate | Primary namespace | Crosslink namespaces | Page type | Action | Rationale | |---|---|---|---|---|---| | External Hermes import methodology | `pixi-vault` | `hermes-agent`, `agent-workflows` | synthesis / source pointer | Crosslink only | Reusable import/routing logic belongs to compiler/source methodology, while this page remains the Hermes-specific assessment. | | Hermes commands/config/providers/profiles/gateway/cron/skills | `hermes-agent` | `agent-workflows`, `eval-trace` | concept or synthesis | Import on demand | Durable only when verified against official docs/local source and tied to Jamie/Pixoid operations. | | Subagents & delegation mechanics | `hermes-agent` | `agent-workflows` | concept / decision table | Crosslink now; import on demand | Hermes owns tool mechanics; Agent Workflows owns when to delegate/spawn/schedule. | | Profile/peer/worker boundaries | `hermes-agent` | `agent-workflows` | concept | Crosslink now; import on demand | Existing workflow pages already cover route boundaries; import mechanics only if configuring profiles. | | Memory providers comparison | `hermes-agent` | `agent-workflows`, `local-ai-infrastructure` | synthesis | Import on demand | High-reuse for provider choice, but stale/risky without official/local verification. | | Deployment backends comparison | `hermes-agent` | `local-ai-infrastructure`, `agent-workflows` | synthesis | Import on demand | Backend tradeoffs are reusable when choosing local/VPS/Docker/remote execution; no active decision in this review. | | Local Models / Airplane Mode | `local-ai-infrastructure` | `hermes-agent`, `agent-workflows` | concept / source pointer | Crosslink only now | Local/offline tradeoffs belong to local infrastructure; Hermes Agent should hold only verified setup mechanics. | | Auxiliary models / side-task routing | `hermes-agent` | `eval-trace`, `agent-workflows` | concept | Import on demand | Useful for evidence-gate and side-task failures, but should be verified against docs/source before trusted. | | Cron troubleshooting checklist | `hermes-agent` | `agent-workflows` | concept/checklist | Import on demand | Likely reusable for crew cron, but best created from a real failure or official-doc check. | | Hermes ecosystem directory triage | `hermes-agent` | `pixi-vault`, `agent-workflows`, `local-ai-infrastructure` | summary | Import on demand | Directory is broad and noisy; only a curated shortlist beats future web search. | | Product/demo uses of Hermes tools | `ai-native-product-surfaces` | `hermes-agent`, `agent-workflows` | source pointer | Defer | New namespace exists, but no candidate source is product-surface-specific yet. | | Dataset/fine-tuning uses of Hermes tools | `curated-tuning-datasets` | `local-ai-infrastructure`, `hermes-agent` | source pointer | Defer | No external Hermes content is dataset-corpus-specific except possible future source collection tooling. | | RL/simulation workflow uses of Hermes tools | `rl-sim-labs` | `eval-trace`, `local-ai-infrastructure`, `hermes-agent` | source pointer | Defer | No external Hermes content is RL-sim-specific unless a tool is adopted for Critical Ranger operations. | ### Practical verdict Keep the review as the durable artifact. Do not import new pages until Jamie asks for one of the on-demand operating questions. The populated namespaces make crosslinks more precise, but they do not justify a broader mirror.