Tectonic Agents deploys a team of Hermes-powered learning agents that watch, fix, and answer for your Minecraft servers and Discord community. They remember every incident, every player question, every config quirk — so the answer is faster tomorrow than it was today.
Managed plans from $9/mo. Full agent team on every paid tier.
log → culprit mod
crash-diffper-tenant, persistent
memoryhard action caps
safetyKeep the signal visible
One place for server health, Discord questions, and the activity your team needs to act on.
Turn crashes into causes
The team remembers your stack and recurring failures, so triage starts with context instead of guesswork.
Move at your trust level
Observe, propose, or act. Every operation stays inside your configured trust gate and hard action cap.
Keep the signal visible
One place for server health, Discord questions, and the activity your team needs to act on.
Turn crashes into causes
The team remembers your stack and recurring failures, so triage starts with context instead of guesswork.
Move at your trust level
Observe, propose, or act. Every operation stays inside your configured trust gate and hard action cap.
Runs your Discord + Minecraft / game servers
Discord
Minecraft
Forge
Fabric
NeoForge
CurseForge
Modrinth
Pterodactyl
Anthropic
OpenAI
NVIDIA NIM
OpenRouter
Hermes
Nous Research
Not a stateless chatbot. A team that gets smarter on your server.
Every tenant gets its own isolated Hermes runtime. That is what makes the learning real — not decoration.
Per-tenant persistent memory
The agent remembers YOUR mods, YOUR rules, YOUR recurring crashes — across restarts and sessions. Not a stateless chatbot.
Self-maintaining skills
The curator promotes repeated fixes into reusable skills. The agent that solved a Forge crash once solves it instantly next time.
Managed model routing
Agents run on platform-managed keys with token-priced credits. Pick a model per agent; heavier models spend more, lighter ones less.
Per-agent model selection
Run each agent on the model that fits its job — a cheap small model for Support, a heavyweight for crash-diffs. You pick per agent; credits are priced by what each model actually costs.
Your data stays isolated
Raw data — servers, secrets, chat, memory — is sealed to your own runtime with row-level security. Only privacy-scrubbed, generalized lessons ever reach the shared layer.
Hermes runtime (Nous Research)
Built on a production agent framework with tool-use, delegation, and cross-session memory — not a thin GPT wrapper.
Six specialists, one orchestrator — including a Development Team.
The orchestrator reads every message and routes to the right specialist — or answers directly. Moderation and Developer are opt-in. The Development Team always ships through Code Review. Every action is metered, capped, and trust-gated.
The agent roster is temporarily unavailable.
The live team will appear here when the roster endpoint responds. Nothing is invented in the meantime.
One team learns it — the whole fleet gets it.
Your runtime is isolated — but it doesn’t have to relearn what the fleet already knows. Tectonic Agents is built around a shared, privacy-scrubbed knowledge base of crash signatures, mod facts, and industry-grade how-tos that every tenant’s agents draw on. A platform that compounds: every server’s lessons make every other server’s agents smarter — so the more the fleet runs, the faster the next crash gets solved.
A shared knowledge base
Crash signatures → cause → fix, mod facts, and industry-grade how-tos, curated into one tenant-agnostic brain every agent can read.
Privacy-scrubbed before it’s shared
Tenant-private data never enters the shared layer. Candidate lessons are stripped of identifiers, paths, IPs and world names — generalized to a signature — before anything is promoted.
It compounds as the fleet grows
Every server’s lessons make every other server’s agents smarter. The more the fleet runs, the faster the next crash gets solved.
How the sharing works — the Researcher agent that curates the knowledge base, the privacy scrub, and the two rings of tenant isolation — is laid out on the How-it-works page.
How it worksLive in three steps.
01
Connect a bot
Paste a Discord or Telegram bot token. Envelope-encrypted the moment it lands — we can’t read it.
02
Add a server
Point at your Minecraft server and panel. Java or Bedrock, CurseForge or Modrinth.
03
Agents run
Your team watches, fixes, and answers — inside a trust-gate that allowlists every action.
6
agents on every tier
100%
tenant isolation
30s
live status polling
24/7
watching your server
A geological seam under your whole operation — one runtime, learning, watching, fixing.
Your server, your agent, your data — sealed off.
Each customer runs in a dedicated, unprivileged runtime with zero access to anyone else’s servers or secrets. Three rings of isolation. Provider keys are envelope-encrypted — we can’t read them and neither can another tenant.
Private agent home
A dedicated, unprivileged runtime per tenant. Zero access to anyone else’s servers or secrets.
Trust-gate allowlist
Every action is checked against an allowlist you control — observe, propose, or act.
Container / microVM
Sandboxed execution on paid tiers. Secrets stay envelope-encrypted end to end.
Priced so a runaway agent can’t surprise you.
Full agent team on every tier. Pay for usage, not features.
Live pricing is temporarily unavailable.
Compare the full tier details on the pricing page once the public pricing endpoint is available.
Open pricingCredits are token-priced (GPT-5.6 Luna = 1× baseline; pricier models cost proportionally more credits). Starter includes 10 GB temporary NVMe scratch for agent work. Hard action cap + hard-stop on every tier.
Straight answers from the product.
The same facts as the rest of this page — memory, trust, credits, and what depends on a live endpoint.
What does Tectonic Agents do?
Tectonic Agents deploys a team of Hermes-powered learning agents that watch, fix, and answer for your Minecraft servers and Discord community. They remember incidents, player questions, and config quirks — so the answer is faster tomorrow than it was today.
How does tenant memory and privacy work?
Each tenant gets its own isolated Hermes runtime and persistent memory. Raw data — servers, secrets, chat, memory — stays sealed to that runtime with row-level security. Only privacy-scrubbed, generalized lessons ever reach the shared layer; identifiers, paths, IPs, and world names are stripped first.
What are trust-gated actions?
Every action is checked against an allowlist you control — observe, propose, or act. Operations stay inside that trust gate, are metered, and are hard-capped so a runaway agent cannot surprise you.
How are pricing and credits presented?
Managed plans start from $9/mo, with the full agent team on every paid tier. Credits are token-priced: heavier models spend more, lighter ones less, and every tier has a hard action cap plus a hard-stop. Live dollar amounts and included credit bundles appear only when the public pricing endpoint responds — this page does not invent rates.
What is live here versus endpoint-dependent?
This landing page is live. The agent roster and pricing cards come from public endpoints and show an honest unavailable state if those endpoints do not respond — nothing is invented in the meantime. Signup versus waitlist follows the published public config.
Deploy a team that learns your server.
Managed plans from $9/mo. Full agent team on every paid tier. Cancel anytime.