Skip to content

modes.md

Latest commit

 

History

History
79 lines (63 loc) · 3.89 KB

modes.md

File metadata and controls

79 lines (63 loc) · 3.89 KB
Raw

Modes — the cost/quality posture toggle

A mode is one word that reassigns the coding role chains (who plans, who executes, who researches) and sets a hard spend ceiling. Switch live with hm mode <name> — no restart. hm mode --list prints the table; the active mode and today's spend show in hm status and the hm dev cockpit.

The two profiles map onto the two headline modes (free, full). The other four are real, supported variations — here for when you want them.

MODE            COST/MO   GPU?  POSTURE
free            $0.00     yes   Kimi-K2.6-free plans, local executes. DEFAULT. Best with Thor/Spark.
full-local      ~$1.50    yes   V4-Pro plans, local executes. V4-Pro judgment over Kimi-free.
full            ~$17      no    V4-Pro plans, V4-Flash executes. No GPU. ~10% of Code Max.
frontier-local  ~$45      yes   Opus plans, local executes. Sovereign + true frontier planning.
frontier        ~$60      no    Opus plans, V4-Flash executes. Closest to Claude Code. Hard sessions.
local           $0.00     yes   Pure local, no API. Air-gapped floor.

Honest framing

  • free and full-local are the headline value — near-frontier planning (Kimi-free or V4-Pro) plus free/sovereign local execution, $0–1.50/month. This is the whole point of owning the GPU.
  • full is the no-GPU on-ramp — anyone can run it, ~$17/mo, ~10% of Claude Code Max.
  • frontier / frontier-local are real but their value narrows versus a Claude Code subscription. They exist for unlimited-usage, no-rate-limit, and private-execution reasons — not pure cost. Use frontier for a genuinely hard session, then drop back to full-local.
  • local is the air-gapped floor: no API at all, limited to local model quality.

The role chains

Each mode declares an ordered provider.model chain per role; a rung whose API key is absent is skipped silently. The roles:

Role What it does
code_plan write the PLAN.md (the expensive, high-leverage step)
code_execute the Hermes loop itself — implement the plan, every turn
code_steer frequent cheap nudges / small corrections
code_repair targeted fix on a verify failure
research_fanout many small parallel calls (query-expand, filter, extract)
research_synth one large synthesis call over the evidence set
code_frontier the rare Opus escalation rung (empty unless a frontier mode)

A role a mode doesn't list falls back to the base chain in config/roles.yaml. The full mode definitions live in config/modes.yaml — edit that file (or your ~/.hermes-max/modes.yaml copy), never the code.

Safety & fall-through

  • A requires_gpu mode (e.g. free) warns rather than silently paying when no local vLLM is reachable — you must explicitly choose to pay (hm mode full).
  • Missing a key drops only that rung: free without OPENROUTER_API_KEY falls through to local planning automatically — no error, no cost.
  • The spend ceiling (local < free < full < frontier) is enforced by the router, so a mode is both a chain swap and a cost cap.

The agent-loop backend swap

A mode also decides which model the Hermes loop itself runs on (distinct from the conductor's per-role routing). hm mode <name> runs scripts/set_mode.sh, which atomically rewrites the model: block of ~/.hermes/config.yaml:

  • local-executor modes (free / full-local / frontier-local / local) → local vLLM ($VLLM_BASE_URL, no key);
  • remote-executor modes (full / frontier) → DeepSeek V4-Flash via the funded DeepInfra endpoint.

It backs up to config.yaml.bak and captures the original model: block once, so the swap is fully reversible (hm mode free rewrites the loop back to local). Skip the live swap with HM_NO_HERMES_SWAP=1.

Deeper detail: architecture.md §11.