Template v1.0 2026-09-12 — As of Hermes Agent v0.21.2

Content / Hermes Agent Releases

September 2026 v0.21.2 – v0.21.4 3 releases

Hermes Agent v0.21.4 (2026.9.21) 2026-09-21

  • ADD--format stream-json — structured JSONL output for a scripted hermes chat -q run, one event object per stdout line
  • ADDskills.auto_load settings key — pin named skills to load into every new session's prompt
  • ADDhermes sessions set-journal-mode — convert state.db's SQLite journal mode after every Hermes process for the profile stops
  • ADDmcp.discovery_concurrency settings key (unverified) — named in the release notes as a configurable MCP discovery connect cap, not yet in the reference docs

Hermes Agent v0.21.3 (2026.9.14) 2026-09-14

  • ADD/loop (alias /proactive) — re-run a prompt on a recurring interval within a session, with --times/--until and status/pause/resume/stop subcommands
  • ADDdisplay.vim_mode settings key — CLI-only vi/vim keybindings for the input composer
  • CHG/queue — bare invocation now lists the pending queue; added list, edit, rm, move and clear subcommands, plus a force-queue add
  • CHG/model — added a --reasoning <level> flag and OpenRouter account-preset selection (@preset/<slug>)
  • CHGhermes auth — OpenRouter credentials can now be added via OAuth browser login (PKCE), alongside the existing API-key flow
  • CHGhermes backup — exclusions expanded to cover regenerable models/, runtimes/ and node/ downloads, more of cache/, and Unix sockets, devices and symlinks

Hermes Agent v0.21.2 (2026.9.11) 2026-09-11

  • ADDcard-keys — initial reference: TUI keybindings plus the transcript, status-bar and overlay symbol glossary (17 rows)
  • ADDcard-slash-core — initial reference: session, context, coding-workflow and run-control commands (29 rows)
  • ADDcard-slash-tools — initial reference: model and reasoning control, tools, skills and profile commands (28 rows)
  • ADDcard-cli — initial reference: global options and every in-scope subcommand (45 rows)
  • ADDcard-settings — initial reference: config.yaml precedence, the seven terminal backends, agent behaviour and limits (22 rows)
  • ADDcard-env — initial reference: HERMES_HOME, secrets handling and backend variables (6 rows)
  • ADDcard-skills — initial reference: the context-file precedence chain, memory, skills and subagent delegation (15 rows)
  • ADDcard-hooks — initial reference: cron, heartbeats, goals, suggestions, blueprints, kanban and MCP (8 rows)

Template / Structure

  • ADDv1.0 — template derived from claude-code-placemat v1.2
  • ADDScope note beside the legend — this placemat covers the CLI and TUI coding surface only; messaging, voice, pets, dashboard, computer use, curator and egress are out of scope
  • CHGPer-agent localStorage keys (pm-hermes-seen-version, pm-hermes-collapsed); theme and density stay shared across every placemat on this origin