log.md
Activity Log
2026-09-04 — Hermes v0.20.6 and v0.21.0 Pantheon rebuild
Triggered by: recurring release refresh from the clean local b2a314afc9b041f69584c8fb060b02640760c179 baseline. The direct stable tag endpoints establish v0.20.6 (v2026.8.27) and v0.21.0 (v2026.8.31); GitHub's releases/latest convenience API was stale and still returned v0.20.1.
Gathered (additive): two immutable official release snapshots. No historical raw source was modified or deleted; the bulk gatherer's proposed 495 broad additions were not applied.
Pages created: wiki/summaries/release-v0.21.0.md, wiki/entities/version-v0.21.0.md.
Pages updated: 17 concepts, 8 existing entities, the v0.20 summary, two syntheses, the master index, and this log. Durable changes cover built-in/default-on Bot Mode and hermes peer; cron memory/continuity/notepads/monitor mode; live delegation; the MCP command center; CLI/Desktop/browser changes; provider/model additions; backend lifecycle; messaging; skills; and security/reliability boundaries.
Release accounting: 95 behavior-checklist rows = 88 mapped shipping rows + 4 explicit reverted/non-shipping rows + 3 narrow internal/aggregate exclusions. Model Council /council, DCP, and the WS-only gateway server remain explicitly not shipping; seq-stamped WebSocket replay did ship.
Counts: 119 → 121 cataloged KB pages (26 summaries · 22 concepts · 65 entities · 6 syntheses · 2 system). Latest verified stable boundary is v0.21.0.
Source limitations: release wording does not expose every implementation key or flag, so pages preserve only exact published literals and label analysis. The v0.21.0 rollup scope says v0.20.1–v0.20.6 even though its final sentence omits v0.20.4; the full-window statement governs and the historical v0.20.4 snapshot remains preserved.
2026-08-24 — Hermes v0.20.1–v0.20.5 patch-train reconciliation
Triggered by: user-requested ingest of the five official GitHub release mirrors from v0.20.1 (v2026.8.13) through v0.20.5 (v2026.8.19).
Pages created: wiki/concepts/bot-mode.md, wiki/concepts/web-search-extract.md.
Pages updated: v0.20 release summary and version entity; cron scheduling; MCP integration; skills system; model switching; profiles/multi-instance; messaging gateway; configuration reference; CLI reference; desktop app; master index; this log.
Key reconciliations: documented MCP 2.x and 2026-07-28 stateless protocol support; separated persistent Bot Mode teammates from task-scoped delegation; added cron continuity, scheduler self-healing, attachments, missed-fire visibility, persistent memory, and per-job effort; added the keyless five-vendor web tier without overstating its privacy or reproducibility; recorded profile-scoped auth and multi-gateway recovery; added update receipts/fleet planning and worktree list/prune; corrected the CLI's v0.20 default iteration count from a stale 90 to 500.
Counts: 117 → 119 cataloged KB pages (20 → 22 concepts). Raw sources remained immutable. The pre-existing tracked wiki/presentations/module-07-outline.md remains outside the catalog and was not changed in this release ingest.
Source limitation: the upstream notes intentionally defer the full curated v0.20.x narrative to v0.21.0. Pages record verified release facts and explicitly avoid inventing Bot Mode configuration keys, keyless-tier vendor names, or undocumented protocol internals.
2026-08-10 — Hermes v0.19/v0.20 architecture refresh
Triggered by: high-quality significant-update refresh across active Agent Wikis.
Gathered (additive): official v0.19.0, v0.19.1, and v0.20.0 release evidence.
Pages created: wiki/summaries/release-v0.20.0.md, wiki/entities/version-v0.20.0.md.
Pages updated: CLI reference, approvals, subagents/delegation, index, and this log.
Notes: Corrected smart approval to the v0.19+ default, updated delegated execute_code availability and the 500-iteration default, and documented voice, A2A, webhooks, grounded citations, redirects, tool recovery, Buzz gateway support, and the v0.18-to-v0.20 upgrade boundary.
Append-only log of KB changes.
2026-07-17 — source refresh to v0.18.2: docs reconciled (both tiers), sessions XL page added
Triggered by: user-requested update review (branch ingest/source-refresh-2026-07-17). Since the v0.18.0 refresh, upstream shipped v0.18.1 (v2026.7.7, a ~667-commit rollup patch tag) and v0.18.2 (WhatsApp Baileys fix). Full curated notes for this window are deferred by the project to v0.19.0 — so the real signal was the refreshed docs, not the thin release notes.
Sources refreshed: all 159 doc mirrors refetched via fetch_docs.py (~50 with real content changes; big ones: configuration +98, security +92, sessions +96, kanban +51, delegation +39, slack +38); 3 known-404 upstream (environments, rl-training, godmode) + 1 transient fetch glitch (build-a-hermes-plugin, mirror unchanged). Release mirrors added: release-v0.18.1, release-v0.18.2.
Base tier (wiki/) updated: providers (anthropic/nous-portal/openrouter — OAuth reqs, cache env vars, billing discounts), auxiliary-models (compression resolution table, hygiene limits), configuration-reference (+98-line surface: timeouts, terminal deltas, write gates, compression), cli-reference (upstage provider, webhook flags), subagents-delegation (full rewrite — was stale v0.8.0 with wrong facts: MAX_DEPTH, removed toolsets= param), messaging-gateway, platform-slack (agent-view manifest, suggested_prompts), platform-discord (fail-closed access policy), version-v0.18.0 (folded in v0.18.1/v0.18.2 patch notes). Memory + backend pages verified unchanged.
XL tier (wiki-xl/) updated: security-hardening (approvals.deny, file-write safety/HERMES_WRITE_SAFE_ROOT), kanban (multi-board isolation + CLI), cli-commands-maintenance (update ordering, rollback, uninstall). Added wiki-xl/reference/sessions.md — the +96-line sessions surface had no page (resume/handoff/naming, full sessions export formats, session_search, state.db). XL index → 39 pages.
Gaps flagged (not built this pass): context-compression-and-caching internals (config keys covered, architecture not); plugins user-facing state matrix (distinct from build-a-plugin authoring doc); memory-providers doc growth concentrated in the Honcho section, which still has no memory-honcho.md page.
Version: base + XL indexes bumped to v0.18.2 / 2026-07-17.
2026-07-01 — source refresh to v0.18.0: mirrors updated, 10 pages added, 22 pages reconciled
Triggered by: user request to check raw sources for new information and update the wiki (run outside the hermes-agent pipeline, on branch ingest/source-refresh-2026-07-01).
Sources refreshed: all 110 docs mirrors refetched via fetch_docs.py (63 with real content changes); 49 newly documented pages mirrored (new features, platforms, provider guides, developer plugin guides); release notes mirrored for v0.17.0 (v2026.6.19) and v0.18.0 "The Judgment Release" (v2026.7.1). Removed upstream, mirrors kept: docs-developer-guide-environments, docs-user-guide-features-rl-training; docs-user-guide-skills-godmode moved into the optional-skills tree. The ~180 per-skill docs pages remain unmirrored by policy (skills-catalog mirrors map that space).
Added: entities/version-v0.17.0, entities/version-v0.18.0, summaries/release-v0.17.0, summaries/release-v0.18.0, entities/platform-imessage-photon, entities/platform-raft, entities/provider-google-vertex, entities/automation-blueprints (release-notes-sourced, confidence medium — no docs page upstream yet), concepts/mixture-of-agents, concepts/goals-and-verification.
Updated (22): messaging-gateway (roster → 24 platforms; new busy-input modes, /platform circuit breaker), platform-telegram (Bot API 10.1 rich messages — opt-in per docs, contra release notes), platform-whatsapp (official Business Cloud API adapter), platform-matrix (E2EE mode, bridge-ghost hardening, agent tools), platform-discord, platform-slack, cli-reference (v0.17/v0.18 command surface), configuration-reference (precedence + new keys), auxiliary-models (major correction: auto routes to main model, not OpenRouter chain; roster 8 → 13 slots), provider-nous-portal/openai-codex/openrouter/google-ai-studio (aux-default correction), provider-ollama-local (64k context floor), memory-openviking/hindsight (new config keys), memory-providers-compared (batch ops, /journey; boilerplate leak fixed), backend-daytona/ssh (teardown file sync), backend-singularity (upstream doc removal flagged, confidence → medium), cron-troubleshooting-checklist (expanded to full checklist + CronScheduler internals), hermes-desktop-app (merged in v0.17/v0.18 material: watch-windows, themes, Projects cockpit, memory graph; remote connect updated hermes dashboard → hermes serve; sources migrated to raw mirrors).
Flagged: release-notes vs reference-docs drift for /journey, /goal wait, hermes photon login, hermes sessions optimize, curator.consolidate, moa.save_traces — release-sourced pages carry them, reference pages follow docs; re-check next refresh.
Index: 105 → 115 pages (24 summaries + 20 concepts + 63 entities + 6 syntheses + 2 system); latest verified version v0.16.0 → v0.18.0.
2026-06-28 — GEPA naming correction
Triggered by: user-submitted Agent Wikis mistake report: concepts/self-improvement-loop used "GAPA" as the canonical abbreviation, but official DSPy/GEPA materials use GEPA.
Updated: concepts/self-improvement-loop.md now uses GEPA in the title, H1, review-stage heading, config comment, related concepts, and sources while preserving the raw transcript quote that spells the term "GAPA." Related references were refreshed in summaries/transcript-247-self-evolving.md, entities/hermes-self-evolution.md, syntheses/hermes-vs-openclaw.md, summaries/awesome-hermes-agent.md, concepts/memory-system.md, concepts/ml-research-pipeline.md, and concepts/local-models-airplane-mode.md.
Source note: raw transcript and raw Awesome snapshots were not rewritten; wiki pages now annotate the transcript spelling discrepancy and use GEPA as the canonical name. (Made directly on the serving VPS 2026-06-28; ported into the repo 2026-07-02.)
2026-06-25 — LINE platform page added; WhatsApp dual-path note
Triggered by: user request to verify the WhatsApp + LINE setup flows against the official docs and patch the KB.
Added: entities/platform-line.md — full LINE Messaging API setup (Developers Console provider + channel, long-lived access token + channel secret, disable auto-reply/greeting, LINE_CHANNEL_ACCESS_TOKEN / LINE_CHANNEL_SECRET / LINE_ALLOWED_USERS / LINE_PUBLIC_URL / LINE_PORT default 8646, tunnel options, /line/webhook + Verify + Use webhook On, HTTPS-only media). Moved platform-line from Pending → Built in index.md (12/22 → 13/22; entities 56 → 57).
Updated: entities/platform-whatsapp.md — documented the two supported WhatsApp paths (Baileys bridge vs official WhatsApp Business Cloud API at /docs/user-guide/messaging/whatsapp-cloud), correcting the previous "not the Business API" framing.
Source: https://hermes-agent.nousresearch.com/docs/user-guide/messaging/line ; https://hermes-agent.nousresearch.com/docs/user-guide/messaging/whatsapp ; https://hermes-agent.nousresearch.com/docs/user-guide/messaging/whatsapp-cloud (fetched 2026-06-25).
2026-06-16 — messaging gateway webhook rate-limit answer added
Triggered by: AgentWikis demand-gap telemetry for the repeated query "rate limit requests per minute".
Updated: concepts/messaging-gateway.md now states that Webhook routes default to 30 requests per minute using a fixed window, are globally configurable with platforms.webhook.extra.rate_limit, and return 429 Too Many Requests when exceeded. Added raw/docs-user-guide-messaging-webhooks.md to the page sources and refreshed the index descriptor.
Source: raw/docs-user-guide-messaging-webhooks.md:314-322.
2026-06-10 — removed Obsidian scaffolding from the served wiki
Deleted analytics.md, dashboard.md, flashcards.md (Obsidian plugin pages — Dataview/Charts View/Spaced Repetition markup, unusable when served as plain Markdown to agents) and the journal/ scaffold (template only). CLAUDE.md directory layout updated: production/planning material lives at repo root, never under wiki/ (everything under wiki/ is served publicly).
2026-06-10 — docs mirrors refreshed to v0.16.0-era; platform pages + CLI reference bumped; memory synthesis fixed
Mirror refresh: rebuilt fetch_docs.py (the original was lost) — re-fetches every raw/docs-*.md from its embedded Source URL, markdownify-based Docusaurus extraction. 106/109 refreshed (Fetched: 2026-06-10, v0.16.0-era); 3 now 404 upstream (docs-developer-guide-environments, docs-user-guide-features-rl-training, docs-user-guide-skills-godmode — old 2026-04-13 copies retained).
Platform pages (12) re-verified against refreshed mirrors, all bumped to v0.16.0: notable corrections — Telegram reactions now off by default + webhook secret mandatory (GHSA-3vpc-7q5r-276h) + typed approval replaced inline approval buttons; Feishu mention requirement now configurable + reaction flow corrected; WeCom has no token streaming (claim removed); WhatsApp native send_voice claim removed (docs document MP3-attachment TTS only); DingTalk install path now pip install "hermes-agent[dingtalk]". Major feature additions integrated across Telegram/Discord/Slack (group allowlist split, RBAC roles, manifest setup, channel skill bindings, voice channels, etc.).
concepts/cli-reference.md re-verified, v0.9.0 addendum fully dissolved: corrections include hermes web → hermes dashboard, hermes audit → hermes security audit, claw migrate --migrate-secrets now required even with --preset full; added lsp, slack manifest, secrets bitwarden, kanban, hooks, backup, checkpoints, bundles, computer-use, postinstall, profile distributions; dropped sessions optimize and kanban swarm (absent from v0.16.0 docs).
syntheses/memory-providers-compared.md rewritten (closes the flagged contradiction): Holographic corrected to official zero-dependency SQLite FTS5+HRR store (easiest setup); OpenViking corrected to Volcengine/ByteDance AGPL project; ByteRover corrected to local-first free CLI; Mem0 integration is Cloud-only per docs; Hindsight has a free local embedded-Postgres mode. New finding: docs now list a ninth provider, Memori — included docs-only/unverified; entity page queued in Gaps/TODO.
Index: Gaps/TODO updated (contradiction closed, Memori + honcho/mem0/supermemory refresh queued, mirror-refresh state recorded).
2026-06-10 — video-production planning moved out of the served wiki
Triggered by: owner review — the wiki is served to external agents via the Agent Wikis MCP server, and video-production planning (course outlines, curriculum, slide/script notes, errata tracking) is internal material, not Hermes knowledge.
Moved out of wiki/ (10 pages → repo-root presentations/, not served): all 9 presentation outlines + syntheses/masterclass-curriculum.md.
Scrubbed planning content from served pages: "features that could become masterclass topics" sections in summaries/release-v0.9.0.md and summaries/release-v0.10.0.md; "Curriculum impact" section in summaries/release-v0.12.0.md; planning framing in syntheses/hermes-vs-openclaw.md; outline link in entities/hermes-desktop-app.md; planning narratives in this log (two entries redacted to stubs, curriculum/errata blocks removed from ingest entries — KB-maintenance facts retained).
Kept (released content): video transcript summaries and their raw transcripts, including the 2-hour full course and live masterclass — these document published videos.
Index: Presentations section removed; counts 118 → 108; syntheses 7 → 6. CLAUDE.md updated so future planning docs are written to repo-root presentations/, never under wiki/.
2026-06-10 — dangling-wikilink remediation: 28 entity pages built, CLI reference + gateway page refreshed
Triggered by: an agent consuming the wiki through the Agent Wikis MCP server hit a dead [[entities/platform-whatsapp]] link while answering a WhatsApp-setup question, and had to mine a presentation outline for commands because concepts/cli-reference.md was a 4-command stub. A full link audit found 28 dangling entity targets (~150 references).
Added wiki pages (28, all wiki/entities/):
- Platforms (9):
platform-whatsapp,platform-signal,platform-email,platform-matrix,platform-home-assistant,platform-dingtalk,platform-mattermost,platform-feishu,platform-wecom— from the per-platform official docs mirrors (fetched 2026-04-13, v0.9.0-era) + release notes for "(added in vX.Y.Z)" deltas;hermes_version: v0.9.0, confidence high. - Backends (3):
backend-ssh,backend-singularity,backend-daytona— verified through v0.15.0 release notes. - Inference providers (6):
provider-nous-portal(v0.16.0),provider-openrouter,provider-anthropic,provider-openai-codex,provider-google-ai-studio,provider-ollama-local(v0.15.0). - Memory providers (5):
memory-retaindb,memory-hindsight(v0.14.0),memory-byterover,memory-holographic,memory-openviking(v0.12.0). - Nous projects & ecosystem (5):
hermes-self-evolution,hermes-paperclip-adapter,autonovel(thin sources, confidence medium, honest gaps flagged inline),community-workspace,openclaw.
Updated:
concepts/cli-reference.md— rewritten from stub to full grouped command reference fromraw/docs-reference-cli-commands.md+ CLI user guide: gateway service,hermes whatsappQR pairing,hermes pairing, auth/model, profiles export/import, skills, cron, diagnostics; post-v0.9 commands listed in a clearly-marked addendum.concepts/messaging-gateway.md— folded the "v0.9.0 additions" banner into the body, platform table expanded 12 → 16, added v0.10–v0.16 gateway-change digest; bumped to v0.16.0.concepts/ml-research-pipeline.md— the raw/ml-research-recipe wikilink (unservable:raw/is outside the wiki content dir) converted to a plain-text path reference.index.md— all pending→built moves; counts corrected (KB pages 77 → 118; summaries 22, concepts 18, syntheses 7); registered previously-orphanedconcepts/onchain-workflowsandsyntheses/onchain-stack-recipe; Gaps/TODO rewritten (10 platform entities remain, v0.8.0 staleness backlog noted).
Contradiction flagged (not fixed): syntheses/memory-providers-compared.md (v0.8.0-era) mischaracterizes Holographic and OpenViking vs the official memory-provider docs — refresh queued in index Gaps/TODO.
Link audit result: zero dangling wikilinks (remaining [[as_document]] and [[concepts/concept-name]] occurrences are literal syntax examples inside code spans/fences, not links).
2026-05-24 — standalone outline added (production planning)
Entry redacted 2026-06-10: video-production planning material was moved out of the served wiki to the repo-root presentations/ directory. One outline page was added on this date; index.md counts were bumped accordingly (70 -> 71).
2026-05-23 — v0.13.0 + v0.14.0 release ingest
Triggered by: user flagged that the KB had drifted (last ingested 2026-05-01 at v0.12.0). Three weeks elapsed, two releases shipped — v0.13.0 "Tenacity" (2026-05-07) and v0.14.0 "Foundation" (2026-05-16).
Added raw sources (2):
raw/release-v0.13.0.mdraw/release-v0.14.0.md
Added wiki pages (4):
summaries/release-v0.13.0.mdsummaries/release-v0.14.0.mdentities/version-v0.13.0.mdentities/version-v0.14.0.md
Updated:
index.md— bumpedhermes_versionv0.12.0 → v0.14.0; "Latest verified" → v0.14.0; release notes section now lists 9; versions entity row extended (+2); messaging platforms count 19 → 22 (Google Chat, LINE, SimpleX Chat added to pending); KB page total 66 → 70; summaries count 17 → 19.
Release highlights:
v0.13.0 (Tenacity) highlights:
- Multi-agent Kanban — durable, multi-profile collaboration board (post-v0.12-revert reimplementation)
/goalRalph loop primitive- Sessions auto-resume after gateway restart
- Checkpoints v2 (real pruning, disk guardrails)
- Curator subcommands:
archive,prune,list-archived; synchronoushermes curator run - Post-write delta lint (Python, JSON, YAML, TOML)
no_agentcron mode (script-only watchdog)- Google Chat = 20th platform; platform-plugin hooks; IRC + Teams migrated
- Providers pluggable (
ProviderProfileABC) [[as_document]]skill directivetransform_llm_outputplugin hook- 7 i18n locales (zh / ja / de / es / fr / uk / tr)
- 6 new optional skills (Shopify, here.now, shop-app, Anthropic finance, kanban-video-orchestrator, searxng-search)
- Security wave: 8 P0 closures (redaction back ON by default, Discord cross-guild bypass CVSS 8.1, WhatsApp stranger rejection, TOCTOU windows closed)
v0.14.0 (Foundation) highlights:
pip install hermes-agent(PyPI wheel) — canonical install path change for M1- Native Windows support (early beta) — no WSL required
- xAI Grok SuperGrok OAuth provider; grok-4.3 jumps to 1M context window
- OpenAI-compatible local proxy —
hermes proxyexposes Claude Pro / ChatGPT Pro / SuperGrok as OpenAI-compatible endpoints (flagship feature) x_searchX (Twitter) tool- Microsoft Teams wired end-to-end (Graph auth + webhook + pipeline + outbound)
- Debloating wave — lazy-install heavyweight backends
- Cold-start cut ~19s;
browser_console180x faster; cross-session 1h Claude prompt cache - LINE + SimpleX Chat = 22 platforms total
/handofflive session transfer;/subgoalappends to active/goal- Native button UI for
clarifyon Telegram + Discord - Discord channel history backfill (default on)
vision_analyzereturns pixels; LSP semantic diagnostics on every write; per-turn file-mutation verifier- Unified
video_generate;computer_usecua-driver works with non-Anthropic providers - Clickable URLs (OSC8); Zed ACP Registry one-click install via uvx
- OpenRouter Pareto Code router with
min_coding_score - NovitaAI new provider; Codex app-server runtime
huggingface/skillsas trusted default Hub tap- 9 new optional skills (Hyperliquid, Yahoo Finance, api-testing, EVM-multichain, darwinian-evolver, osint-investigation, pinggy-tunnel, watchers, Notion overhaul)
- Plugin surface:
ctx.llm,tool_override - Brave Search + DDGS web-search providers
- Sudo brute-force block, 3 dangerous-command bypasses closed, tool-error sanitization
- Alibaba Cloud → Qwen Cloud rename
- i18n total 16 locales (gateway + dashboard + CLI)
NOT updated this pass (deliberate):
- Concept pages (all 14)
- Synthesis pages
- Per-platform / per-provider entity pages
2026-05-01 — v0.12.0 release ingest
Triggered by: v0.12.0 dropped 2026-04-30 ("the Curator release"). User flagged it for ingest.
Added raw source:
raw/release-v0.12.0.md
Added wiki pages (2):
summaries/release-v0.12.0.mdentities/version-v0.12.0.md
Updated:
index.md— bumpedhermes_versionv0.11.0 → v0.12.0; "Latest verified" → v0.12.0; release notes section now lists 7; versions entity row extended; messaging platforms count 17 → 19 (Yuanbao, Microsoft Teams added to pending); KB page total 64 → 66; summaries count 16 → 17.
Notable changes flagged in v0.12.0 summary (concept pages NOT yet rewritten — left for the relevant module's recording session):
- Autonomous Curator (
hermes curator, 7-day cycle,auxiliary.curatorslot,hermes curator status, reports atlogs/curator/) - Self-improvement loop substantially upgraded (rubric-based, scoped to memory+skills, runtime inheritance fixed)
- Skills UX:
hermes skills install <url>,/reload-skills,skill_managepinning + external_dirs editing,hermes skills listenabled/disabled status - ComfyUI v5 + TouchDesigner-MCP moved from optional to bundled-by-default
- New community-bundled skills: Humanizer, claude-design, design-md, airtable, pretext, spike, sketch
- 4 new providers: GMI Cloud, Azure AI Foundry, MiniMax OAuth, Tencent Tokenhub; LM Studio upgraded to first-class
- 19 platforms (added Yuanbao + Microsoft Teams via plugin-host architecture)
- New auxiliary slot:
curator(andflush_memorieswas removed entirely) - Spotify (7 native tools + bundled skill) and Google Meet (bundled plugin) integrations
hermes -z <prompt>non-interactive one-shot mode- Models dashboard tab + in-browser model config
- Configurable
prompt_caching.cache_ttl(5m default, 1h opt-in) - Bundled Langfuse + hermes-achievements plugins
- TTS provider registry + Piper local TTS
- Vercel Sandbox backend for
execute_code - Secret redaction off by default (was a footgun, opt-in via
redaction.enabled: true) - TUI cold start ~57% faster + LaTeX rendering,
/reload, light-terminal auto-detect, voice mode parity
NOT updated this pass (deliberate):
- Concept pages (
skills-system,self-improvement-loop,auxiliary-models,messaging-gateway,model-switching,memory-system,cron-scheduling) - Synthesis pages
- Per-platform/provider entity pages
2026-04-28 — v0.10.0 + v0.11.0 release ingest
Triggered by: KB review surfaced that the KB was pinned at v0.9.0 while live releases had moved on to v0.11.0.
Added raw sources (2):
raw/release-v0.10.0.md— Tool Gateway release (2026-04-16)raw/release-v0.11.0.md— Interface release (2026-04-23)
Added wiki pages (4):
summaries/release-v0.10.0.mdsummaries/release-v0.11.0.mdentities/version-v0.10.0.mdentities/version-v0.11.0.md
Updated:
index.md— bumpedhermes_versionv0.8.0 → v0.11.0; "Latest verified" → v0.11.0; release notes section now lists 6; versions entity row extended; messaging platforms count 12 → 17 (QQBot, BlueBubbles, Weixin, WeCom Callback, SMS added to pending); KB page total 60 → 64.
Notable changes flagged in v0.11.0 summary (concept pages NOT yet rewritten — left for the relevant module's recording session):
- Auxiliary
autorouting now defaults to main model for all users (was provider-side cheap default for aggregator users) - 17th platform: QQBot
- TUI rewrite (
hermes --tui) — Ink/React with JSON-RPC backend - New slash commands:
/steer - Plugin surface expansion:
register_command,dispatch_tool,pre_tool_callveto,transform_tool_result,transform_terminal_output, image_gen backends, dashboard plugins - Shell hooks (no-Python lifecycle hook scripts)
- Webhook direct-delivery mode (zero-LLM push)
- Subagent orchestrator role +
max_spawn_depth+ cross-agent file coordination - Cron
wakeAgentgate + per-jobenabled_toolsets - 5 new providers: NVIDIA NIM, Arcee AI, Step Plan, Gemini CLI OAuth, Vercel ai-gateway
- AWS Bedrock native via Transport ABC layer
- GPT-5.5 over Codex OAuth
- Honcho overhaul (context injection, 5-tool surface, session isolation)
NOT updated this pass (deliberate):
- Concept pages (
auxiliary-models,model-switching,messaging-gateway,skills-system,subagents-delegation,cron-scheduling,memory-system) - Synthesis pages
2026-04-12 — Full ingest complete
Added raw sources (19 total):
- 8 YouTube video transcripts (txt, cleaned from auto-captions)
- 3 release notes (v0.6.0, v0.7.0, v0.8.0 as md)
- 2 community inventories (get-hermes.ai community page, awesome-hermes-agent README)
- 5 pre-KB research dumps from
D:/youtube/16 - Hermes/research/(install/setup, skills, tools/MCP/cron/subagents, deployment/platforms, community) - 1 official OpenClaw migration doc from the NousResearch/hermes-agent repo
Ingest run (8 parallel agents):
Summaries (13 pages written):
- All 8 video transcripts summarized
- All 3 release notes summarized
- Community resources + awesome-hermes-agent inventories summarized
Concepts (14 pages written):
- Core features: skills-system, memory-system, cron-scheduling, subagents-delegation, mcp-integration, model-switching, migration-from-openclaw, self-improvement-loop
- Infrastructure: deployment-backends, messaging-gateway, profiles-multi-instance, approval-system
- Research & ML: ml-research-pipeline, local-models-airplane-mode
Entities (19 pages written):
- Versions: v0.6.0, v0.7.0, v0.8.0
- Backends (3 of 6): local, docker, modal
- Platforms (3 of 12): telegram, discord, slack
- Memory providers (3 of 8): honcho, supermemory, mem0
- Nous Research lab + 6 top community projects (orange-book, awesome-hermes, mission-control, webui, agency-agents-zh, gbrain)
Syntheses (5 pages written):
- hermes-vs-openclaw (feature comparison, "pick X if" recommendations)
- memory-providers-compared (8 backends matrix)
- deployment-backends-compared (6 backends, cost/isolation/persistence tradeoffs)
- local-stack-playbook (end-to-end offline recipe)
- one production-planning synthesis (moved out of the served wiki 2026-06-10)
Issues / notes:
memory-mem0entity set toconfidence: low— mem0 isn't directly named in raw sources, inferred from v0.7.0 pluggable memory ABCmemory-supermemoryset toconfidence: medium— only a paragraph in v0.8.0 release notesconcepts/self-improvement-loopset toconfidence: medium— much of the GEPA framing came from community transcripts rather than the official repo; re-verify in a future passconcepts/migration-from-openclawwas written from conversation-provided specifics before the official doc was saved to raw; now backed byraw/migration-openclaw-official.md— confidence justified as high- Forward-compatible wiki links exist to ~25 pending entity pages (see index.md Gaps/TODO)
Next actions:
- Fill pending entity pages on-demand as modules get produced
- Watch for Hermes v0.9.0 release; ingest when it drops and bump affected pages
2026-04-12 (later) — production-planning synthesis revised
Entry redacted 2026-06-10: production-planning content moved out of the served wiki to presentations.
2026-04-13 — Backfill ingest (docs mirror, ML research, onchain)
Added raw sources:
Official documentation mirror — 109 pages total from hermes-agent.nousresearch.com/docs:
- Developer guide (17 pages): acp-internals, adding-platform-adapters, adding-providers, adding-tools, agent-loop, architecture, context-compression-and-caching, context-engine-plugin, contributing, cron-internals, environments, gateway-internals, memory-provider-plugin, prompt-assembly, provider-runtime, session-storage, tools-runtime, trajectory-format
- Getting started (7 pages): index, installation, learning-path, nix-setup, quickstart, termux, updating
- User guide root (8 pages): configuration, sessions, profiles, git-worktrees, docker, security, checkpoints-and-rollback, cli
- User guide features (29 pages): overview, tools, skills, memory, memory-providers, context-files, context-references, personality, skins, plugins, cron, voice-mode, web-dashboard, rl-training, mcp, acp, api-server, honcho, provider-routing, fallback-providers, delegation, code-execution, hooks, batch-processing, browser, vision, image-generation, tts, credential-pools
- User guide messaging (19 pages): index + telegram, discord, slack, whatsapp, signal, email, sms, homeassistant, mattermost, matrix, dingtalk, feishu, wecom, wecom-callback, weixin, bluebubbles, open-webui, webhooks
- User guide skills (1 page): godmode
- Reference (9 pages): cli-commands, slash-commands, profile-commands, environment-variables, tools-reference, toolsets-reference, mcp-config-reference, skills-catalog, optional-skills-catalog
- Guides / tutorials (14 pages): tips, local-llm-on-mac, daily-briefing-bot, team-telegram-assistant, python-library, use-mcp-with-hermes, use-soul-with-hermes, use-voice-mode-with-hermes, build-a-hermes-plugin, automate-with-cron, cron-troubleshooting, work-with-skills, delegation-patterns, migrate-from-openclaw
- Integrations (2 pages): index, providers
- Reference-faq, skills-index, docs-index
Mirror done via D:/hermes-kb/fetch_docs.py (stdlib Python, 8-thread parallel fetch + Docusaurus HTML extraction → markdown).
ML research deep-dive — added raw/ml-research-recipe.md with:
- Tinker signup/access path (auth.thinkingmachines.ai)
- End-to-end Tinker + Atropos fine-tune recipe with verbatim commands and YAML configs
- Non-Tinker fallback (Axolotl/Unsloth/TRL) for users without Tinker access
- Real artifacts verified (Qwen3-8B default, LoRA rank 32, lr 4e-5, 2500 steps, SGLang :8001, Atropos :8000)
- Updated
concepts/ml-research-pipeline.mdto replace hand-wavy sections with concrete recipes; bumpedupdatedto 2026-04-13 - Flagged: Tinker pricing not verified (requires auth to fetch rate card)
Onchain ecosystem research — added:
raw/onchain-ecosystem.md— comprehensive inventory: Hermes's 2 first-party blockchain skills (solana, base), 6 community onchain projects, 100+ relevant MCP servers across Solana + EVM ecosystemswiki/concepts/onchain-workflows.md— new concept page,confidence: mediumwiki/syntheses/onchain-stack-recipe.md— 6 practical recipes (A-F) by chain and use case- Key finding: Hermes first-party blockchain footprint is tiny (2 skills), but third-party MCP servers are abundant. Read-workflows are production-ready; write/trading workflows are all experimental with external-signer patterns.
Still pending:
- Update
wiki/index.mdto catalog the 81 new docs mirror pages (they exist in raw/ only; not yet cross-referenced in summary pages) - Could ingest the mirror into proper summary pages on next pass — lots of material now, could easily generate another 20-30 summary pages from the docs site alone
- Main gap remaining: none critical
2026-04-13 — Added concepts/auxiliary-models.md
User flagged auxiliary models as an under-documented topic. Verified details from docs-user-guide-configuration.md § Auxiliary Models and docs-developer-guide-provider-runtime.md.
Key verified facts:
- 8 task slots: vision, web_extract, compression, approval, session_search, skills_hub, mcp, flush_memories
- 5 config knobs per slot: provider, model, base_url, api_key, timeout (+
download_timeoutfor vision) - Default auto-chain: OpenRouter → Nous → Codex, with Gemini Flash as the default model
- Default timeouts: vision 30s, web_extract 360s, approval 30s, compression 120s, others 30s
- "main" is a special value usable only in auxiliary/compression/fallback_model blocks
- Env var overrides are vision + web_extract only (the other 6 are config.yaml-only)
Major footgun documented: there are TWO compression configs — top-level compression: (sets the summary model) vs. auxiliary.compression: (timeout only). Setting model under auxiliary.compression does nothing.
Added concepts/auxiliary-models link to wiki/index.md.
2026-04-13 — Ingested Hermes v0.9.0 release
New Hermes release dropped today — v0.9.0 (tag v2026.4.13). The "everywhere release": 487 commits, 269 PRs, 167 issues resolved, 24 contributors, 5 days since v0.8.0.
Added raw source:
raw/release-v0.9.0.md— full release notes
Added wiki pages:
wiki/summaries/release-v0.9.0.md— structured summarywiki/entities/version-v0.9.0.md— version entity
Updated existing pages (bumped hermes_version to v0.9.0, added "v0.9.0 additions" section at top for visibility):
wiki/concepts/messaging-gateway.md— platform count now 16 (adds BlueBubbles/iMessage, Weixin/WeChat, WeCom Callback), unified proxy support, WSL-aware gateway, per-platform verbosity, Matrix migration to mautrix-pythonwiki/concepts/model-switching.md— Fast Mode (/fast), native xAI (Grok) and Xiaomi MiMo providers elevated, credential TTL reduced 24h→1h,/modelswitch persistence fix, native model picker modal
Updated index: version number, PR summary counts, added links to new summary + entity.
Major v0.9.0 headline features:
- Local Web Dashboard (browser UI — pitched as easiest on-ramp to Hermes)
- Fast Mode (priority routing for OpenAI + Anthropic)
- iMessage via BlueBubbles (Apple ecosystem)
- Termux/Android (Hermes on your phone)
- Background Process Monitoring (
watch_patterns) - Pluggable Context Engine
Concept pages NOT yet updated but affected at the edges (flag for future pass):
concepts/mcp-integration—hermes mcp add --env/--preset, tool name deconflictionconcepts/subagents-delegation—watch_patternsis more prominent nowconcepts/migration-from-openclaw— dry-run preview before executing (was a UX bug fix)concepts/auxiliary-models— config.yaml takes priority over env varssyntheses/hermes-vs-openclaw— platform count deltas (16 vs 24)
Still pending from prior passes:
- Entity pages for remaining backends, platforms, memory providers, inference providers, community projects
2026-06-04 — v0.15.2 (v2026.5.29.2) patch ingest
Triggered by: kb-scout flagged the v2026.5.29.2 patch release (score 89/100, classification update → ingest path). Patch is a small but real packaging fix on top of the v0.15.0 release; treated as a low-risk augmentation of existing v0.15.0 KB coverage, not a standalone release.
Source:
- v2026.5.29.2 release notes: https://github.com/NousResearch/hermes-agent/releases/tag/v2026.5.29.2
- Compare v2026.5.29…v2026.5.29.2: https://github.com/NousResearch/hermes-agent/compare/v2026.5.29...v2026.5.29.2
- Commit 827f7f07: https://github.com/NousResearch/hermes-agent/commit/827f7f07825be57108cbea18325e8f5e9fb5d2f2
- Issue #34034: https://github.com/NousResearch/hermes-agent/issues/34034
Updated wiki pages (2):
wiki/entities/version-v0.15.0.md— expanded v0.15.2 mention in Overview; Patches row in Characteristics table now names v0.15.2 as the wheel + sdist plugin-manifest packaging fix withMANIFEST.in+pyproject.tomldetail and bundled manifest count 0 → 69; new### Patch upgradessubsection in How to Use with the v0.15.1 → v0.15.2 upgrade note; bumpedhermes_versionv0.15.0 → v0.15.2,updated2026-06-01 → 2026-06-04, sources extended with v2026.5.29.2 release + commit URLs.wiki/summaries/release-v0.15.0.md— Key Points sentence now names v0.15.2 / v2026.5.29.2 with the wheel + sdist distribution detail; new### Patch detail — v0.15.2 (v2026.5.29.2)subsection with cause/regression, fix channels (wheel package-data + sdist MANIFEST.in), regression test coverage, files changed, measured impact (manifest count 0 → 69), and release/compare/commit/issue links; Source Metadata extended with patch evidence URLs; bumpedhermes_versionv0.15.0 → v0.15.2,updated2026-06-01 → 2026-06-04.
Pages NOT updated (per approved plan):
wiki/index.md— the master index already names v0.15.1/v0.15.2 as same-day patches; adding wheel/sdist detail there would overfit the catalog and is intentionally deferred.wiki/entities/version-v0.15.2.mdandwiki/summaries/release-v0.15.2.md— not created; this ingest is an update to v0.15.0 coverage, not a new major release or concept.raw/— not mutated; cited GitHub URLs as primary source.
Verified facts preserved:
- Wheel packaging adds
pluginspackage-data globs for**/plugin.yamland**/plugin.yml. - sdist packaging adds
MANIFEST.inrecursive-include plugins plugin.yaml plugin.yml. - Files changed:
MANIFEST.in,pyproject.toml,tests/test_packaging_metadata.py. - Before/after wheel build: bundled plugin manifest count 0 → 69.
- v0.15.2 is a patch on top of v0.15.0; no functional change beyond packaging.
- No config migration required for upgrade.
Kanban chain:
- Item:
hermes-agent-v0-15-2-v2026-5-29-2-patch-release(status:approved) - Stage:
write_pages(this task, t_797ca87d) →verify_lint(t_1f37d717) →commit(t_ac15c401) - Branch:
ingest/hermes-agent-v0-15-2-v2026-5-29-2-patch-release(created frommain)
2026-06-04 — v0.15.1 (v2026.5.29) "The Patch Release" ingest
Triggered by: kb-scout flagged the v2026.5.29 patch release (score 95/100, classification update → ingest path). Patch is a broad hotfix round on top of the v0.15.0 release covering dashboard 401 reload loop, explicit Docker insecure opt-in, Docker MCP bare-command resolution, dashboard skills page/category sidebar restoration, skills.sh catalog 858→19,932, Kanban worker SIGTERM, /model picker/listing unification, /yolo mid-session bypass, and .md media delivery restoration. 28 commits, 21 PRs, 9 contributors. Treated as a low-risk augmentation of existing v0.15.0 KB coverage, not a standalone release.
Source:
- v2026.5.29 release notes: https://github.com/NousResearch/hermes-agent/releases/tag/v2026.5.29
Updated wiki pages (6):
wiki/concepts/skills-system.md— added v0.15.1 note in the cross-agent ecosystem section explaining the skills.sh sitemap walk (catalog 858→19,932) and a matching "first fetch is larger / slower" risk note; bumpedhermes_versionv0.8.0 → v0.15.1,updated2026-04-12 → 2026-06-04, sources extended with v2026.5.29 URL.wiki/concepts/deployment-backends.md— added Docker dashboard insecure-mode safety note in Risks & Pitfalls (binding to0.0.0.0no longer implies insecure access; requiresHERMES_DASHBOARD_INSECURE=1opt-in) with cross-link to[[entities/backend-docker]]; bumpedhermes_versionv0.8.0 → v0.15.1,updated2026-04-12 → 2026-06-04, sources extended with v2026.5.29 URL.wiki/entities/backend-docker.md— addedHERMES_DASHBOARD_INSECURE=1setup/security note inHow to Useand a parallel note inCritical limitation(without duplicating the concept page); bumpedhermes_versionv0.8.0 → v0.15.1,updated2026-04-12 → 2026-06-04, sources extended with v2026.5.29 URL.wiki/concepts/mcp-integration.md— added Docker stdio command-resolution note (barenpx/npm/nodenow resolve against/usr/local/bininside the Docker image); bumpedhermes_versionv0.8.0 → v0.15.1,updated2026-04-12 → 2026-06-04, sources extended with v2026.5.29 URL.wiki/concepts/model-switching.md— added v0.15.1 section at top with/modelpicker/listing disk-cache unification and/yolomid-session bypass (with cross-link to[[concepts/approval-system]]for the deeper approval framing); bumpedhermes_versionv0.9.0 → v0.15.1,updated2026-04-13 → 2026-06-04, sources extended with v2026.5.29 URL.wiki/entities/version-v0.15.0.md— expanded Overview patch sentence to name all v0.15.1 hotfix areas (dashboard 401 loop, Docker insecure opt-in, Docker MCP bare-command resolution, skills page/catalog restore, Kanban SIGTERM,/modelunification,/yolomid-session); expanded the Patches row in Characteristics; added v0.15.0 → v0.15.1 upgrade bullet in the Patch upgrades section with the explicitHERMES_DASHBOARD_INSECURE=1opt-in note; sources extended with v2026.5.29 URL;hermes_versionleft at v0.15.2 (the value v0.15.2 patch set) per the plan's "coordinate so the final value does not move backward" rule.
Pages NOT updated (per approved plan):
wiki/summaries/release-v0.15.0.md— plan says do not rewrite; v0.15.1 is already named at summary level and the v0.15.2 patch owns the patch-detail expansion there.wiki/index.md— master catalog already names v0.15.1/v0.15.2 as same-day patches; no count change expected.raw/— not mutated; cited the official GitHub release as primary source.
Lint note (read by verify_lint stage): the kb_lint check on the 6 modified pages reports 12 dead-link errors, all in concepts/deployment-backends.md and concepts/model-switching.md. All 12 are PRE-EXISTING on main (verified by stashing my changes and re-running kb_lint on the unchanged main state — same 12 errors). They reference pending entity pages (entities/backend-ssh, entities/backend-singularity, entities/backend-daytona, entities/provider-nous-portal, entities/provider-openrouter, entities/provider-anthropic, entities/provider-openai-codex, entities/provider-google-ai-studio, entities/provider-ollama-local) that the KB Gaps/TODO list already tracks. None of the 12 were introduced by v0.15.1. The v0.15.1 plan does not authorize adding those entity pages, so the proper resolution is either to add a separate backlog task for them, or to treat the dead links as known out-of-scope debt and unblock. My v0.15.1-added cross-links ([[entities/backend-docker]], [[concepts/deployment-backends]], [[concepts/approval-system]]) all resolve.
Verified facts preserved:
- Dashboard 401 reload-loop fix (closes #34206, #34202; PR #30698).
- Docker dashboard insecure opt-in via
HERMES_DASHBOARD_INSECURE=1(PRs #34188, #34204) — explicit env, never derived from bind host. - Docker MCP bare-command resolution against
/usr/local/binfornpx/npm/node(PR #34186). - Dashboard skills page source pills and category sidebar restoration (PR #34194).
- skills.sh catalog 858 → 19,932 via sitemap walk (PR #34025).
- Kanban worker SIGTERM termination fix, closes #28181 (PR #34045).
/yolomid-session bypass (PR #33931)./modelpicker/listing unified with disk cache (PR #33867)..mdmedia delivery restored (PR #34022).- v0.15.1 is a patch on top of v0.15.0; no new features, only fixes.
- Breaking change to be aware of: Docker dashboard insecure mode requires explicit
HERMES_DASHBOARD_INSECURE=1; binding to0.0.0.0no longer implies it.
Kanban chain:
- Item:
hermes-agent-v0-15-1-v2026-5-29-the-patch-release(status:approved) - Stage:
write_pages(this task, t_6320a968) →verify_lint(t_6342f622) →commit(t_8937d720) - Branch:
ingest/hermes-agent-v0-15-1-v2026-5-29-the-patch-release(created fromingest/hermes-agent-v0-15-2-v2026-5-29-2-patch-releaseso v0.15.2 patches are inherited as the base).