Hermes Agent v0.18.0 (v2026.7.1) — The Judgment Release¶
Release Date: July 1, 2026 Since v0.17.0: ~1,720 commits · 998 merged PRs · 2,215 files changed · ~251,000 insertions · ~41,000 deletions · 949 issues closed · 381 community contributors
The Judgment Release. Over the last week and a half the team put nearly all of its effort into one goal: resolve every P0 and P1 issue and PR in the entire Hermes Agent repo — and as of this release, 100% of them are closed. Zero open P0s. Zero open P1s. On top of that clean-sweep, v0.18.0 is about how well Hermes thinks and how it knows when its work is actually done.
🎯 The P0/P1 Clean Sweep — 100% Resolved¶
This is the release headline. Every single P0 and P1 across the whole repo is now closed:
| Priority | Issues Closed | PRs Merged |
|---|---|---|
| P0 (critical) | 3 | 8 |
| P1 (high) | 493 | 188 |
| Total | 496 | 196 |
~692 highest-priority items resolved in twelve days. The team intends to keep P0/P1 at zero from here forward.
✨ Highlights¶
🧠 Mixture-of-Agents — First-Class Model¶
MoA graduated from a mode to a first-class model provider. Every named MoA preset appears as a selectable model under a moa provider alongside Claude, GPT, and Grok in every model picker (CLI, TUI, desktop, gateway). Pick "my-council" the same way you'd pick any model.
- See every model's reasoning — each reference model's full output renders as its own labelled block
- Aggregator streams live — watch the answer form in real time instead of waiting for the whole response
- Reference models see full tool state and fire on every user/tool response
- Opt-in full-turn trace persistence to JSONL for debugging and eval
✅ Verification & Goals — The Agent Proves Its Work¶
Hermes now records verification evidence and can decide it's finished by actually running your project's checks, not by claiming success.
- Completion contracts for
/goal— state what "done" looks like; the loop judges completion against evidence, not the model's say-so - Coding verification evidence ledger — profile-scoped record of canonical project checks
pre_verifyhook for wiring in custom verification/goal wait <pid>— park the standing-goal loop on a background process
🎓 Self-Improvement: /learn & /journey¶
/learn <anything>— distill a reusable skill from a directory, URL, or workflow you just walked through. Honors CONTRIBUTING.md skill standards automatically./journey— CLI + TUI learning timeline of accumulated memories and skills, with in-place edit/delete- Memory graph in desktop — playable radial timeline of memories + skills over time
- Cheaper background review — aux-model routing + context digest + adaptive cadence for the post-turn self-improvement fork
🖥️ Desktop Coding Cockpit¶
The desktop app gained real, per-profile Projects — a sidebar of your codebases, a coding rail, a review pane, git worktree management, and agent-facing project tools.
- First-class Projects with
project → repo → lanemodel - Multi-terminal panel with read-only agent terminals; persist & restore across relaunch
- PR-style file diffs in chat; in-app spot editor for file previews
- Conversation timeline rail for long threads
- Composer refactor wave — god-file de-entanglement into isolated engine hooks
- Pets — roaming pet (opt-in), pop-out overlay, notifications
🤖 Background Subagent Fan-Out¶
delegate_task can now fan out multiple subagents that all run in the background: your chat is never blocked, and when every subagent finishes, their results come back as a single consolidated turn. Kick off "research these five competitors in parallel" and carry on with something else.
🌐 Scale-to-Zero Gateway¶
The gateway can go dormant when idle and quiesce cleanly before a restart, migration, or auto-update — without dropping in-flight conversations. External drain coordination ensures nobody gets cut off mid-turn. Running Hermes at scale for a team is now production-grade.
🔌 Google Vertex AI Provider¶
Vertex AI is now a first-class provider for Gemini over the OpenAI-compatible endpoint. Hermes auto-mints and refreshes short-lived OAuth2 tokens from a service-account JSON or Application Default Credentials — no static key, no mid-session expiry.
✍️ /prompt — Compose in Your Editor¶
/prompt opens your $EDITOR so you can hand-write a long, multi-line prompt in real markdown instead of fighting a one-line input box.
🔒 Security Hardening¶
- MCP-config persistence attack surface locked down
- Cron
base_urloverrides that could exfiltrate provider credentials blocked - Non-reusable sentinel for prefix secrets in file reads
- Slack
xapp-token redaction - Browser cloud-metadata floor enforced on every backend
aiohttpCVE floor across lazy messaging paths
🖥️ Desktop App Details¶
Coding Cockpit¶
| Feature | Description |
|---|---|
| Projects sidebar | Per-profile codebases with project → repo → lane model |
| Multi-terminal panel | Read-only agent terminals, persist & restore across relaunch |
| PR-style diffs | File diffs rendered in chat with inline annotations |
| Spot editor | Edit files directly in the preview pane |
| Git cockpit | Status, review, worktrees from the desktop |
| Conversation timeline | Navigate long threads with a visual rail |
UX & Surfaces¶
| Feature | Description |
|---|---|
| Read aloud (TTS) | Auto-TTS toggle on the composer |
| Window state persistence | Size, position, maximized remembered across launches |
| Context-usage popover | See where your context budget is going |
| Shared overlay Panel | Reused across cron, profiles, agents |
| Floating composer | Pop the input into a draggable window |
| Backup management | Import/create/download from web UI |
🧠 Core Agent & Architecture¶
Delegation & Subagents¶
- Background fan-out — parallel subagents run in background, one consolidated return when all finish
- Background subagent tracking in CLI + TUI status bar
Agent Loop & Tools¶
- One-shot LLM helper +
llm.oneshotgateway RPC - Coding-context project facts exposed via
project.factsRPC web_extracttruncate-and-store instead of LLM summarization- Friendly human-phrased tool labels for built-in tools
/reasoning full— uncapped thinking/timestamps+ timestamps in/history- Multiple
HERMES_WRITE_SAFE_ROOTdirectories
Compression & Sessions¶
- In-place compaction option (single session ID)
- Backup includes
projects.dband kanban boards in pre-update snapshot
🌐 Gateway, Fleet & Relay¶
Scale-to-Zero & Drain¶
- Idle detection + dormant-quiesce (Phase 0)
- External drain coordination (Phase 2) — safe-shutdown without dropping conversations
- Persist in-flight transcript on restart/shutdown drain timeout
- Self-heal gateway stranded in draining/degraded state
Relay (Phase 5/6)¶
- Wake primitive (gateway side)
- Going-idle / buffered-flip primitive
- Multi-platform-per-agent identity
- Authorize relay-delivered events by delivery, not source platform
Gateway Core¶
- Typed send-error classification (
SendResult.error_kind) - Per-platform
typing_indicatortoggle - Per-category context breakdown in
/usage - Configurable concurrent-run cap to prevent DoS
📱 Messaging Platforms¶
| Platform | New in v0.18.0 |
|---|---|
| Telegram | Configurable command menu, rich draft preview gating, drain send pool |
| Slack | Opt-in Block Kit rendering, --no-assistant manifest flag |
| Discord | Reasoning rendered as -# subtext via display.reasoning_style |
| Native media delivery via Baileys bridge | |
| Signal | AAC voice-note remux, shared markdown formatting |
| Teams | Native send_video/send_voice/send_document |
| All | Cron continuations — thread-preferred with DM-mirror fallback |
🔧 Tool System, Skills & MCP¶
- Blank Slate setup mode — minimal agent, opt in to everything
- MCP config persistence attack surface hardened;
base_urlexfil blocked - Kanban: task lifecycle plugin hooks (claimed/completed/blocked); typed block reasons
- LSP: PowerShellEditorServices language server
- mem0: v3 API + OSS mode + update/delete tools
cloudflare-temporary-deployoptional skill- Creative-ideation v2.1.0 method library
⚡ Performance¶
| Improvement | Impact |
|---|---|
| Lazy-load gateway platform adapters | Faster cold start |
libyaml CSafeLoader for config/plugins |
Parse speed boost |
FTS5 segment merge + handoff_state index |
Reduced SQLite write-lock contention |
Single-pass list_profiles + skill-count cache |
Faster profile enumeration |
👥 Contributors¶
381 people contributed to this release — the largest cohort yet.
Core¶
- @teknium1 — release lead; MoA first-class, verification/goals,
/learn, background review, security, providers, P0/P1 sweep - @OutThisLife — desktop app (Projects, memory graph,
/journey, multi-terminal, composer refactor, pets, verification UX) - @kshitijk4poor — P0/P1 backlog burn: cron reliability, state perf, security, gateway
- @benbarclay — relay Phase 5/6, scale-to-zero/drain, dashboard auth/keys
- @ethernet8023 — CI/docker (unified jobs, faster builds)
- @helix4u — Windows hardening
🔁 Notable Reverts¶
- Cron job storage returned to per-profile
auth.jsoncloning removed (duplicating OAuth grant causes sibling revocation)- Windows terminal-popup PRs rolled back
prompt_caching.enabledtoggle backed out for re-evaluation
📊 Release Stats¶
| Metric | Value |
|---|---|
| Commits | ~1,720 |
| Merged PRs | 998 |
| Files Changed | 2,215 |
| Insertions | ~251,000 |
| Deletions | ~41,000 |
| Issues Closed | 949 |
| Contributors | 381 |
| P0/P1 Open at Release | 0 |
← v0.17.0 — The Reach Release | Changelog Home →
This Hermes repo is one of the largest structured collections of public AI, automation, business, and technology documentation. Content remains attributed to original authors and repositories. Indexed and organized by www.CorpusIQ.io.