Skip to content

solo-developer.md

Latest commit

 

History

History
93 lines (67 loc) · 4.17 KB

solo-developer.md

File metadata and controls

93 lines (67 loc) · 4.17 KB

Reference Architecture: Solo Developer

VPS + phone bot + cost routing. The 80% setup — cheap, reliable, reachable anywhere, good enough for one person's real work.

Who this is for

  • Developer / maker who wants a Telegram/Discord driver for their daily work
  • No ops team; "set it and forget it" is the design goal
  • Willing to spend $5–50/mo to not run your own hardware

Cost

  • Infra: $5–7/mo (Hetzner CX22 or Fly machine)
  • LLM: $20–60/mo for typical personal use with cost routing
  • Domain + DNS: $0–1/mo

Total: ~$25–70/mo.

Architecture

  phone/laptop                Internet             hermes.yourdomain.com
       │                         │                         │
       └── Telegram/Discord ─────┼── Cloudflare/Caddy ────→│
                                 │                         │
                                 │                         ├── hermes.service
                                 │                         ├── hermes-dashboard.service
                                 │                         ├── Langfuse (self-host)
                                 │                         └── LightRAG
                                 │
                                 └── Anthropic / Google / Moonshot / Cerebras

Parts list

  • Hetzner CX22 (Debian 12 or Ubuntu 24.04) — $5/mo, 4GB RAM, 2 vCPU
  • Domain ($12/yr) — or use a free subdomain from duckdns/nip.io
  • Telegram bot token (free; @BotFather)
  • API keys: Anthropic (default), Google (Gemini Flash for triage), optionally Moonshot + Cerebras for coding/classification

Install

# As root on a fresh VPS
curl -sSL https://raw.githubusercontent.com/OnlyTerp/hermes-optimization-guide/main/scripts/vps-bootstrap.sh | bash

Then:

# As root
sudo -u hermes nano ~/.hermes/.env          # fill in keys
sudo cp /opt/hermes-optimization-guide/templates/config/cost-optimized.yaml \
     /home/hermes/.hermes/config.yaml       # or telegram-bot.yaml
sudo cp /etc/caddy/Caddyfile.hermes.reference /etc/caddy/Caddyfile
# edit /etc/caddy/Caddyfile — replace *.yourdomain.com
sudo systemctl reload caddy
sudo systemctl start hermes hermes-dashboard

Why cost-optimized.yaml is the right default

See templates/config/cost-optimized.yaml. Defaults to Gemini Flash (cheapest smart model), uses Cerebras Llama for classification (near-free), and only escalates to Sonnet for high-stakes coding. With prompt caching + Fast Mode disabled by default, typical cost is $0.05–0.30 per active hour.

If you need max quality for a specific task, just say "use sonnet" in chat — the router honors explicit user overrides.

Routines you'll run

Every one of these is installed by the bootstrap script (symlinks into ~/.hermes/skills/):

  • Morning: /cost-report window=24h — yesterday's spend
  • On idle threads: /telegram-triage (autoreply)
  • Weekly: /weekly-dep-audit severity_floor=high
  • Nightly: /nightly-backup s3://my-backups/hermes/ 30 (or set remote=local if you don't care)

Scaling ceilings

Constraint Hit at Fix
CX22 RAM ~5–10 concurrent tool calls + LightRAG Upgrade to CX32 ($12/mo)
Gemini Flash free tier 1500 req/day Route to Cerebras or add paid quota
LightRAG on 2 vCPU Indexing 10MB+ docs Move indexing to a spot Modal sandbox
Cost budget $50+/mo Turn on prefer_cached: true + 32K compression trigger

Security note

Because this box is public-facing, always deploy the denylist + require_approval from cost-optimized.yaml, and keep your Telegram bot private (restrict allowed_user_ids to your own ID). Any "public" bot should use a separate token and run in a quarantine profile — see Part 19 and the security-hardened.yaml template.

When to graduate