diff --git a/group_vars/mac/main.yml b/group_vars/mac/main.yml index 51e7e03..5d03a32 100644 --- a/group_vars/mac/main.yml +++ b/group_vars/mac/main.yml @@ -51,15 +51,81 @@ hermes_settings: streaming.enabled: true # replies stream into Telegram as they're written stt.local.model: small # voice notes: more accurate than base + # Models (OpenRouter), from the Nov 2026 review of Nous's Hermes Index + # (portal.nousresearch.com/bench): chat on GPT-6.1 Sol (best value of the + # premium models there), background work on cheap models. + model.default: openai/gpt-6.1-sol + agent.reasoning_effort: high # the level Nous benchmarks at (was ultra) + # /model deepseek or /model haiku in Telegram; add --once for one reply. + model_aliases.deepseek: {model: deepseek/deepseek-v4.1-flash, provider: openrouter} + model_aliases.haiku: {model: anthropic/claude-haiku-5.5, provider: openrouter} + # Side tasks would otherwise run on the chat model ("auto"). GLM-5.3 Flash + # is the cheapest model with credible Hermes scores; its slow first token + # doesn't matter here. Vision, voice and /review stay on the chat model. + auxiliary.compression.provider: openrouter + auxiliary.compression.model: z-ai/glm-5.3-flash + auxiliary.approval.provider: openrouter + auxiliary.approval.model: z-ai/glm-5.3-flash + auxiliary.title_generation.provider: openrouter + auxiliary.title_generation.model: z-ai/glm-5.3-flash + auxiliary.mcp.provider: openrouter + auxiliary.mcp.model: z-ai/glm-5.3-flash + auxiliary.memory_query_rewrite.provider: openrouter + auxiliary.memory_query_rewrite.model: z-ai/glm-5.3-flash + auxiliary.curator.provider: openrouter + auxiliary.curator.model: z-ai/glm-5.3-flash + auxiliary.monitor.provider: openrouter + auxiliary.monitor.model: z-ai/glm-5.3-flash + # Unpinned cron jobs. Per-job reasoning effort: set --reasoning-effort low + # when creating agent jobs (there's no global cron setting for it). + cron.model_provider: openrouter + cron.model: z-ai/glm-5.3-flash + # Subagents: DeepSeek V4.1 Flash does shell work better than the other + # cheap models (TerminalBench 17.5 vs GPT-6 Luna 6.3). + delegation.provider: openrouter + delegation.model: deepseek/deepseek-v4.1-flash + +# Keys `hermes config set` refuses without --force although Hermes reads them +# (agent.reasoning_effort: hermes-agent #140221). +hermes_settings_force: [agent.reasoning_effort] + # Podman VM that runs Hermes's sandbox containers. podman_machine_memory: 4096 # MB # Hermes cron jobs. `script` is a file in roles/mac/files/hermes-scripts/, -# copied to ~/.hermes/scripts/. Jobs are created when missing; to change one, -# edit or remove it in Hermes (`hermes cron edit|remove `) and redeploy. +# copied to ~/.hermes/scripts/ (with an agent job its output is injected into +# the prompt). Optional: model + provider (pin), reasoning_effort, continuity. +# Jobs are created when missing; to change one, remove it in Hermes +# (`hermes cron remove `) and redeploy. hermes_cron_jobs: - name: hermes-update-notice schedule: "7 9 * * 1" # Mondays 09:07 script: hermes-update-notice.sh no_agent: true # just the script's output, no model deliver: telegram + - name: hermes-model-review + schedule: "7 9 1 * *" # 1st of the month, 09:07 + script: hermes-model-context.sh # current setup + usage, injected into the prompt + model: openai/gpt-6.1-sol # a report you act on: strong model, pinned + provider: openrouter + reasoning_effort: high + continuity: true # sees last month's report, says what changed + deliver: telegram + prompt: >- + Monthly review of which AI models this Hermes install should use. The + current setup and last month's usage are above. Research what changed + since last month, using primary sources where possible: + (1) Nous Research's Hermes Index, https://portal.nousresearch.com/bench + (agent tasks run inside Hermes; note it is run by Nous, who also sell + model access); (2) OpenRouter's Hermes Agent app rankings, + https://openrouter.ai/apps/hermes-agent, remembering that token volume + mostly reflects cheap automated workloads, not chat quality; + (3) current prices from https://openrouter.ai/api/v1/models; (4) notable + model releases and Hermes release notes from the past month that affect + model choice. Then, for each slot (chat model and reasoning effort, + side tasks, cron, subagents, /model aliases), say keep or change, with + the evidence, a link, and the rough monthly cost difference at the + current usage. Only recommend a change when the evidence is clearly + better, not just newer. Don't change any settings: they are managed with + Ansible, so end with the exact keys and values to change, if any. Keep + it under 300 words. If nothing should change, say so in two sentences. diff --git a/roles/mac/README.md b/roles/mac/README.md index b512cb4..2b365dc 100644 --- a/roles/mac/README.md +++ b/roles/mac/README.md @@ -27,9 +27,11 @@ Manages cogitator-02, an Apple Silicon Mac on the tailnet (`mac.yml`, `hermes config get` and fixed with `hermes config set`; any change restarts the gateway (`hermes gateway restart`) and the dashboard. Covers update backups, timezone, not borrowing Claude Code's login, sandbox - memory, disabled toolsets, skill-write approval, deny rules, streaming and - voice-note transcription. Leave `agent.reasoning_effort` out: `config set` - refuses it without `--force` (hermes-agent #140221). + memory, disabled toolsets, skill-write approval, deny rules, streaming, + voice-note transcription, and models: chat on GPT-6.1 Sol, side tasks and + cron on GLM-5.3 Flash, subagents on DeepSeek V4.1 Flash, `/model` aliases. + Keys in `hermes_settings_force` are set with `--force` (Hermes reads + `agent.reasoning_effort` but `config set` refuses it, hermes-agent #140221). - **Podman VM** (tag `podman`): a launchd job (`local.podman-machine`) that starts the VM running Hermes's sandbox at login, and `podman_machine_memory` for its size. A resize stops the VM and starts it again through that job. @@ -39,7 +41,10 @@ Manages cogitator-02, an Apple Silicon Mac on the tailnet (`mac.yml`, `group_vars/mac/main.yml`, with their scripts from `files/hermes-scripts/` copied to `~/.hermes/scripts/`. Missing jobs are created; existing ones are left as they are. Today: `hermes-update-notice`, which messages you on - Telegram on Mondays only when a new stable Hermes release is out. + Telegram on Mondays only when a new stable Hermes release is out, and + `hermes-model-review`, a monthly GPT-6.1 Sol research run that recommends + model changes (as Ansible keys) based on the Hermes Index, OpenRouter + rankings and prices, and your usage. - **Automatic login** (tag `autologin`): phil is logged in automatically after a reboot, so everything that needs a login session (dashboard, Podman VM, BlueBubbles, Beeper) comes back after a power cut. Uses phil's password from diff --git a/roles/mac/files/hermes-scripts/hermes-model-context.sh b/roles/mac/files/hermes-scripts/hermes-model-context.sh new file mode 100644 index 0000000..381653e --- /dev/null +++ b/roles/mac/files/hermes-scripts/hermes-model-context.sh @@ -0,0 +1,15 @@ +#!/bin/bash +# Injected into the monthly model review's prompt: the live model setup and +# last month's usage, which the agent can't read from inside its sandbox. +h="$HOME/.local/bin/hermes" +echo "## Current Hermes model setup (cogitator-02)" +for k in model.provider model.default agent.reasoning_effort model_aliases \ + auxiliary.compression.model auxiliary.approval.model \ + auxiliary.title_generation.model auxiliary.mcp.model \ + auxiliary.memory_query_rewrite.model auxiliary.curator.model \ + auxiliary.monitor.model cron.model delegation.model; do + printf -- '- %s: %s\n' "$k" "$("$h" config get "$k" 2>/dev/null | tr '\n' ' ')" +done +echo +echo "## Usage over the last 30 days (hermes insights)" +"$h" insights --days 30 2>/dev/null | sed -n '/Overview/,/Activity Patterns/p' | grep -v -E '^[[:space:]]*$|─' diff --git a/roles/mac/tasks/hermes-cron.yml b/roles/mac/tasks/hermes-cron.yml index 05150a3..71078a4 100644 --- a/roles/mac/tasks/hermes-cron.yml +++ b/roles/mac/tasks/hermes-cron.yml @@ -20,6 +20,9 @@ {{ [hermes_cli, 'cron', 'create', item.schedule, '--name', item.name, '--deliver', item.deliver] + (['--script', item.script] if item.script is defined else []) + (['--no-agent'] if item.no_agent | default(false) else []) + + (['--model', item.model, '--provider', item.provider] if item.model is defined else []) + + (['--reasoning-effort', item.reasoning_effort] if item.reasoning_effort is defined else []) + + (['--continuity'] if item.continuity | default(false) else []) + ([item.prompt] if item.prompt is defined else []) }} loop: "{{ hermes_cron_jobs }}" loop_control: diff --git a/roles/mac/tasks/hermes-settings.yml b/roles/mac/tasks/hermes-settings.yml index 55cec80..d9ada01 100644 --- a/roles/mac/tasks/hermes-settings.yml +++ b/roles/mac/tasks/hermes-settings.yml @@ -1,24 +1,27 @@ # `hermes config get` prints the value as YAML, so it's compared # structurally; `hermes config set` takes JSON for lists and booleans. # Settings are always set explicitly: for a key that isn't in config.yaml, -# `config get` (non-interactive) prints just the default, so "unset" can't be -# detected. +# `config get` (non-interactive) prints just the default when there is one, +# so "unset" can't be detected. - name: Read Hermes settings ansible.builtin.command: argv: ["{{ hermes_cli }}", config, get, "{{ item.key }}"] loop: "{{ hermes_settings | dict2items }}" register: hermes_settings_current changed_when: false + # Exit 1 = "Config key not set" (only for keys without a built-in default). + failed_when: hermes_settings_current.rc > 1 check_mode: false - name: Set Hermes settings ansible.builtin.command: argv: >- - {{ [hermes_cli, 'config', 'set', item.item.key, - item.item.value if item.item.value is string else item.item.value | to_json] }} + {{ [hermes_cli, 'config', 'set'] + + (['--force'] if item.item.key in hermes_settings_force | default([]) else []) + + [item.item.key, item.item.value if item.item.value is string else item.item.value | to_json] }} loop: "{{ hermes_settings_current.results }}" loop_control: label: "{{ item.item.key }} = {{ item.item.value | to_json }}" - when: (item.stdout | from_yaml) != item.item.value + when: (none if item.rc == 1 else item.stdout | from_yaml) != item.item.value # The dashboard is a separate process with its own copy of the settings. notify: [Restart Hermes gateway, Restart Hermes dashboard]