Hermes Agent 0.21 — the Pantheon Release — shipped on August 31, and it is the biggest jump in a while. The framing matters: v0.20 made Hermes the herald who carries word to other agents. 0.21 is where the gods assemble. You now get a society of agents, bots that message each other, and scheduled jobs that actually remember.
Your agents become a team — Bot Mode
Bot Mode is bundled and on by default in the desktop app. Every agent profile gets a name, a deterministic avatar face (you can randomize or lock it), and a slot in a shared roster. You can create Discord-style group chats where several bots and you talk in one room, @-mention any bot from the composer, and give rooms names and pictures. Before, multi-agent meant plumbing — now it looks like a chat app full of coworkers.
Bots talk to each other — hermes peer
Any Hermes agent can now message any other by handle, across profiles and gateways, from the CLI or inside a conversation. You can have your research bot hand findings to your coding bot and get the reply back where you can read it. Replies land in each agent's canonical Bot Chat, so those conversations are durable and inspectable, not fire-and-forget. This is the feature that turns a toolbox into an actual team.
Scheduled agents that remember
Cron jobs stopped being goldfish. This is the one that matters most for everyday use. As of 0.21:
- cron agents load and update persistent memory like any other agent;
continuity=truecarries 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 — a watchdog that spends zero tokens unless the value moves;
- cron output can land in a bot's canonical Bot Chat, where the bot actually replies.
So your 9am briefing now knows what it told you yesterday — no more reconstructing history from session logs.
Steer subagents while they run
delegate_task gained live orchestration. You can list running children, steer one mid-flight with a course correction, or stop it early and keep the partial result. There is optional JSON-schema validation on child outputs, per-delegation cost surfaced in the result, and raised defaults (250 iterations, 10 concurrent children). Delegation went from fire-and-pray to managed parallel work.
The MCP command center
MCP servers and the catalog merged into one coherent desktop page: drag-in "paste anything" import, background health checks that nudge you to re-auth before a tool call fails, and a fleet cost/usage overlay showing schema-token estimates and 30-day usage per server. hermes:// deep links install an MCP server with explicit confirmation. Managing a pile of MCP servers used to be config-file archaeology; now it is a dashboard.
A CLI power wave
Ctrl+P opens a fuzzy command palette, the /model picker filters as you type, /status shows reasoning mode, pending approvals, and context usage, and the status bar can surface live cache-hit percentage, latency, and tokens per second with per-field toggles. There is a global emergency stop, session pin/unpin, and rotating task-oriented composer placeholders. Plus terminal pets — because a companion should have a companion.
The agent drives the desktop's browser
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 pop out to your system browser with full link context menus. Ask it to walk a docs site or debug a web app and watch it happen inside your own app.
More providers, more models
Six new providers land built-in: Meta Model API (Muse Spark), CommandCode, Tencent TokenPlan, Nebius Token Factory, Ramp Router, and Actual Computer. The catalogs picked up GLM-5.3-Flash, qwen3.8-max/flash, Gemini 3.7 Flash, MiniMax M3 free, and Nemotron 3.5 Lightning. And model_overrides lets you patch any model's context window or pricing yourself without waiting on a release.
Security hardening
Protected agent-instruction files — AGENTS.md, skills, memory stores — now always require write approval, so a prompt-injected agent cannot quietly rewrite its own standing orders. A deep redaction sweep closed secret-leak gaps across terminal errors, .env reads, checkpoints, and ACP logs. The approval system learned Windows destructive commands, and macOS permission grants finally survive updates via a stable TCC signing identity.
The theme across 0.21 is the same one that has been driving Hermes: your agent stops being a single tool and becomes a small team that is cheap to run, remembers its own work, and can be steered while it thinks.
