Hermes Agent v0.21.0: The Pantheon Release

Nous Research published Hermes Agent v0.21.0 on August 31, 2026 as GitHub tag v2026.8.31. The changelog nicknames it The Pantheon Release, and it is unusually direct about the product bet.

v0.20.0 made Hermes the herald — he spoke, and he carried word to other agents. In v0.21.0 the gods assemble.

Bot Mode now ships built into the desktop app: a society of named agents with their own faces and group chats, where bots talk to each other — and to you — like a team, not a toolbox. Around that spine, cron jobs gained memory so scheduled agents actually learn between runs, subagents can be steered live mid-flight, the MCP surface became a command center, and the agent can drive the desktop's own browser. This tag documents the v0.20.1–v0.20.6 infrastructure patch windows; those rollups are not the story here.

Since v0.20.0 the notes claim about 5,800 commits, 2,475 merged PRs, 5,680 files changed, roughly 869,000 insertions and 135,000 deletions, about 2,100 issues closed, and 760+ contributors. @teknium1 is credited with release direction.

Bot Mode, bundled and default-on

Bot Mode is a bundled, default-on part of the desktop app. Every agent profile gets a name, a deterministic avatar face (with randomize and lock controls), and a place in a shared roster. You can create Discord-style group chats where multiple bots and you talk in one room, @-mention any bot from the composer, and give rooms names and pictures. Before, the notes say, "multi-agent" meant plumbing; now it looks like a chat app full of coworkers.

Highlights credit @teknium1, @OutThisLife, and @dokterdok. Desktop follow-through in the same window includes attributed agent-to-agent message cards, sender-side delivery notices, paint-first hydration for instant wakes, editable group names and room pictures, and a Routines pane.

hermes peer: durable bot-to-bot DMs

Any Hermes agent can now message any other by handle, across profiles and gateways, from the CLI or from inside a conversation. Replies land in each agent's canonical Bot Chat, so conversations between agents are durable and inspectable rather than fire-and-forget. The notes' own example: ask a research bot to hand findings to a coding bot and get the reply back where you can read it.

Cron jobs that remember

Scheduled jobs, the changelog says, stopped being goldfish. Cron agents now load and update persistent memory like every other agent. continuity=true carries each run's output into the next, so a monitor can dedupe against what it already reported. Every job gets a durable notepad scratchpad. Monitor-mode jobs skip the LLM entirely when nothing changed. Cron output can land in a bot's canonical Bot Chat, where the bot actually responds. A 9am briefing job now knows what it told you yesterday.

Steer subagents while they run

delegate_task gained live orchestration: list running children, steer one mid-flight with a course correction, or stop it early and keep the partial result. Optional JSON-schema validation lands on child outputs, per-delegation cost is surfaced in results, and defaults rose to 250 iterations and 10 concurrent children. Delegation, the notes say, went from fire-and-pray to managed parallel work.

MCP command center and an agent-driven browser

MCP servers and the catalog merged into one desktop page with drag-in "paste anything" import, background health checks that nudge you to re-auth before a tool call fails, a fleet cost/usage overlay showing schema token estimates and 30-day usage per server, and hermes:// deep links that install an MCP server with explicit confirmation. Managing twenty MCP servers used to be config-file archaeology; now it is a dashboard.

The in-app browser stopped being a window the agent could only look at. Hermes now navigates, clicks, and reads it directly, and pages can be popped out to the system browser with full link context menus.

Standing orders, secrets, Windows approvals

Writes to protected agent-instruction files — AGENTS.md, skills, and memory stores — always require approval, so a prompt-injected agent cannot quietly rewrite its own standing orders. A redaction sweep closed secret-leak gaps across terminal errors, .env file reads, checkpoints, and ACP logs. The approval system learned Windows destructive commands. macOS permission grants survive updates via a stable TCC signing identity.

Also in the tag — and what did not ship

A CLI power wave landed alongside the pantheon features: Ctrl+P fuzzy command palette, a /model picker that filters as you type, /status showing reasoning mode, pending approvals, and context usage, live status-bar metrics (cache-hit percent, latency, tokens/sec), a global emergency stop, session pin/unpin, and terminal pets.

Six new providers arrived: Meta Model API (Muse Spark), CommandCode, Tencent TokenPlan, Nebius Token Factory, Ramp Router, and Actual Computer. Catalog additions include GLM-5.3-Flash, qwen3.8-max and qwen3.8-flash, Gemini 3.7 Flash, MiniMax M3 free, and Nemotron 3.5 Lightning. model_overrides lets you patch a model's context window or pricing without waiting on a release.

Explicitly not shipping after land-then-revert: Model Council mode (/council), the DCP context engine, and a WS-only gateway server. FastAPI remains on the desktop boot path. Seq-stamped event replay did ship.

@teknium1 is credited with about 1,340 merged PRs in this window. Core-team volume also lists @kshitijk4poor (374 PRs) and @OutThisLife (209 PRs). The notes thank 760+ contributors.

If you already run Hermes desktop, skip the model-catalog tour and try the product shift: name two Bot Mode profiles, start a group chat, send a hermes peer DM, and confirm the thread lands in canonical Bot Chat. Then read the v0.21.0 / v2026.8.31 release notes for cron continuity, live delegate_task steering, and the MCP command-center page.

Sources

• Hermes Agent v0.21.0 (v2026.8.31)