✓ SubscribeSubscribers: 4
Blind dev projects news
Nous Research #announcements
@everyone
☤ Hermes Agent v0.18.0 — The Judgment Release
~1,720 commits · 998 PRs · 2,215 files · 949 issues closed · 370+ community contributors — twelve days.
v0.16.0 put Hermes on your desktop. v0.17.0 was about how far it reaches. v0.18.0 is about how well it thinks — and how it knows when its work is actually done. Mixture-of-Agents became a first-class model you can pick, with every reference model's reasoning shown to you and the aggregator's answer streamed live. The agent learned to verify its own work against evidence, @@CODE0@@ gained completion contracts, and @@CODE1@@ + `
◆ Mixture-of-Agents is now a first-class model you can pick. Every named MoA preset shows up as a selectable model under a `moa` provider — pick "my-council" the way you'd pick Claude or GPT, in the CLI, TUI, desktop, or gateway, and Hermes routes your prompt through that ensemble automatically. An ensemble of frontier models deliberating on your hardest questions is one selection away.
◆ @@CODE4@@ — turn anything into a reusable skill by describing it. @@CODE5@@ distills a skill from a directory, a URL, or the workflow you just walked Hermes through — written to your CONTRIBUTING.md standards automatically. Teaching Hermes a new trick is a single command now, not a skill-authoring session.
◆ `
◆ Delegate a pile of work and keep going. `delegate_task` fan-out now runs multiple subagents in the background — chat never blocks — and returns one consolidated summary when they all finish. "Research these five competitors in parallel," then carry on while a small fleet works.
◆ First-class coding Projects in the desktop app. Per-profile Projects — a sidebar of codebases, a coding rail, a review pane, git worktree management, agent project tools — backed by a proper `project → repo → lane` model. The desktop is turning into an actual coding cockpit.
◆ Cheaper self-improvement and web scraping. 60x speed and 50x cheaper web scraping, and the post-turn "learn from what just happened" fork now routes to an auxiliary model, digests context instead of replaying the whole conversation, and adapts its cadence — a fraction of the old cost, same self-improvement.
◆ Google Vertex AI — Gemini through your GCP service account. Vertex is now a first-class provider for Gemini over its OpenAI-compatible endpoint. Vertex has no static API key — every request needs a short-lived OAuth2 token minted from a service-account JSON or ADC, which is why pasting a token into a custom provider always died mid-session. Hermes now mints and auto-refreshes those tokens for you. If your org runs Gemini through Google Cloud, point Hermes at your service account and it just works.
Plus: @@CODE9@@ composes your next message in @@CODE10@@ · `
Update: @@CODE12@@ (or @@CODE13@@).
Full changelog: https://github.com/NousResearch/hermes-agent/releases/tag/v2026.7.1
https://fxtwitter.com/NousResearch/status/2072413332665962617?s=20
@everyone
☤ Hermes Agent v0.18.0 — The Judgment Release
~1,720 commits · 998 PRs · 2,215 files · 949 issues closed · 370+ community contributors — twelve days.
v0.16.0 put Hermes on your desktop. v0.17.0 was about how far it reaches. v0.18.0 is about how well it thinks — and how it knows when its work is actually done. Mixture-of-Agents became a first-class model you can pick, with every reference model's reasoning shown to you and the aggregator's answer streamed live. The agent learned to verify its own work against evidence, @@CODE0@@ gained completion contracts, and @@CODE1@@ + `
/journey` made self-improvement something you can see and steer. Underneath: a gateway you can genuinely run at scale, first-class coding projects on the desktop, and background subagent fan-out.◆ Mixture-of-Agents is now a first-class model you can pick. Every named MoA preset shows up as a selectable model under a `moa` provider — pick "my-council" the way you'd pick Claude or GPT, in the CLI, TUI, desktop, or gateway, and Hermes routes your prompt through that ensemble automatically. An ensemble of frontier models deliberating on your hardest questions is one selection away.
◆ @@CODE4@@ — turn anything into a reusable skill by describing it. @@CODE5@@ distills a skill from a directory, a URL, or the workflow you just walked Hermes through — written to your CONTRIBUTING.md standards automatically. Teaching Hermes a new trick is a single command now, not a skill-authoring session.
◆ `
/journey` + memory graph — see everything Hermes has learned. A learning timeline in the CLI and TUI shows the memories and skills Hermes has accumulated, and you can edit or delete any of them right there. The desktop's new memory graph lays them on a playable radial timeline. For the first time you can see what your agent knows and prune what's wrong.◆ Delegate a pile of work and keep going. `delegate_task` fan-out now runs multiple subagents in the background — chat never blocks — and returns one consolidated summary when they all finish. "Research these five competitors in parallel," then carry on while a small fleet works.
◆ First-class coding Projects in the desktop app. Per-profile Projects — a sidebar of codebases, a coding rail, a review pane, git worktree management, agent project tools — backed by a proper `project → repo → lane` model. The desktop is turning into an actual coding cockpit.
◆ Cheaper self-improvement and web scraping. 60x speed and 50x cheaper web scraping, and the post-turn "learn from what just happened" fork now routes to an auxiliary model, digests context instead of replaying the whole conversation, and adapts its cadence — a fraction of the old cost, same self-improvement.
◆ Google Vertex AI — Gemini through your GCP service account. Vertex is now a first-class provider for Gemini over its OpenAI-compatible endpoint. Vertex has no static API key — every request needs a short-lived OAuth2 token minted from a service-account JSON or ADC, which is why pasting a token into a custom provider always died mid-session. Hermes now mints and auto-refreshes those tokens for you. If your org runs Gemini through Google Cloud, point Hermes at your service account and it just works.
Plus: @@CODE9@@ composes your next message in @@CODE10@@ · `
/reasoning full` shows uncapped thinking · friendly human-phrased tool labels · Krea via managed Nous Subscription · Z.AI endpoint picker · Blank Slate setup mode · a big cron reliability wave · Windows console-flash + gateway-restart hardening · cold-start perf (lazy adapters, libyaml config parsing) · desktop composer refactor wave · roaming pets. 949 issues closed.Update: @@CODE12@@ (or @@CODE13@@).
Full changelog: https://github.com/NousResearch/hermes-agent/releases/tag/v2026.7.1
https://fxtwitter.com/NousResearch/status/2072413332665962617?s=20