Management
One pane of glass.
Every agent, every channel, every task — operated from a single command surface. Talk to your workforce from Telegram, iMessage, WhatsApp, or the web, without ever switching context.
The System
The Pantheon Orchestration System
Pantheon is our agent orchestration system — it lets us deploy and coordinate an entire workforce under a single command surface, while maintaining optimal cost efficiency and task efficacy in every domain.
Cost efficiency
$0 in API calls on most days.
We route every task across our private cloud and your providers — so compute cost approaches zero while quality stays at the frontier.
Custom fit
Our roster is just the starting point.
Start from our battle-tested Agents — then we tune their memory, voice, and toolset to your domain. Need an agent we don't have? We build new ones from the same scaffolding.
Scale at your pace
Start with one. Grow to a workforce.
Deploy a single agent for the workflow that hurts most. Once your team is comfortable, add the next. Pantheon was built to scale from one seat to an entire organization.
Our starter roster
Eight production-ready Agents.
Each one has its own memory, voice, toolset, and domain — and each is a starting template for a custom agent tuned to your business.
☉I
Hermetic_Demiurge
Creator / Builder
Full-stack dev, infra, CI/CD. Creates other Agents.
⛩︎II
Janus
GEO / SEO
Finds every door into your market and opens them — organic, AI answer engines, geo.
⚒︎III
Hephaestus
Engineer / Product
Refactors, upgrades, rebuilds. Turns prototypes into durable product.
ΔIV
Metis
Executive Communication
Exec comms, stakeholder alignment, drafts that close.
♨︎V
Prometheus
Revenue & GTM
Growth, outreach, GEO audits, proposals, pipeline.
☸︎VI
Dzambhala
Wealth / Finance
Investment management. $100K+ AUM, 1%+ daily on crypto, equity, options.
☷VII
Mahakala
Security & Privacy
Testing, privacy, infra hardening, incident response.
𓁟VIII
Thoth
Knowledge & Memory
Docs, hybrid-search memory, cross-Agent knowledge.
We customize from these tried-and-tested archetypes for every engagement. Bring us a function we don't list — we'll build the agent for it.
Engagement
From first call to live workforce.
Four steps. We do the heavy lifting at every one.
-
1
Discovery
We map your team, your work, the loops we'd automate, and where agents earn their keep.
90 minutes · with you -
2
Scope
We propose which agents, which gateways, which integrations, and what "done" looks like in 30 days. You approve.
~3 business days -
3
Deploy
Your private Pantheon goes live on a dedicated cloud, wired to your channels and pre-loaded with your context.
Overnight -
4
Manage
We run the infrastructure, cover the model + compute bills (up to your tier cap), and keep your roster compounding skills weekly.
Ongoing
Deployment
Three ways to run your Pantheon.
Every engagement starts with a fully private, single-tenant deployment. We pick the substrate together.
Private Cloud
A dedicated VPS on the cloud you choose — AWS, GCP, Azure, Hetzner, your call. Fully single-tenant. Your data lives in your infrastructure. We administer.
Available today · Most engagements
On Our Hardware
We host on Prompt Engines' own racks. Same single-tenant guarantees. We manage the metal end-to-end. Useful when you'd rather not manage a cloud-provider relationship.
Available today
Pantheon SaaS
Multi-tenant, self-serve onboarding for smaller teams. Faster, lighter touch, lower commitment. Join the waitlist
Q4 2026 · Waitlist open
Every deployment includes
- Full Hermes runtimeThe orchestration core, plus all 12 archetype agents, ready to be tailored.
- The Pantheon dashboardA single pane of glass for your workforce, sessions, and skills.
- All gateway integrationsTelegram, Slack, Discord, WhatsApp, Email, Signal — meet your team where it works.
- Continuous skill evolutionAgents learn from every task. Skills consolidate weekly without your input.
- Model + infra costs coveredUp to a custom cap set at scoping. We absorb the overage. Reviewed quarterly.
- Security hardeningSecrets rotation, dependency audits, secure config reviews — Mahakala-grade.
- Subpoena-grade audit logsEvery tool call, model call, file touched, message sent — recorded.
- Approval gatesTirith permission layer: four levels, inline approvals, configurable cost ceilings.
- Exit-ready, alwaysHermes is open source. Walk away with config, memory, skills, history — at any time.
Capabilities
What makes Pantheon different.
Not a chat interface. Not a copilot. A self-improving workforce with institutional memory and 40+ tools per agent.
Persistent institutional memory
Three-layer memory: working context, structured files, and vector episodic recall. Agents remember everything across every session.
Self-improving skills
After three task completions, agents auto-create a reusable skill. Skills refine from feedback and compound week over week.
Multi-channel by default
Telegram, Discord, Slack, WhatsApp, Signal, Email — one agent, every channel. Context syncs automatically.
Model-agnostic
Switch between 200+ models via OpenRouter, or run local with Ollama. One interface, any backend, no lock-in.
Subagent delegation
Spawn isolated parallel agents for concurrent workstreams. Shared skill library, isolated context, faster throughput.
Cron-driven execution
Natural-language scheduling: "Every morning at 8am, send me a briefing." Runs unattended, delivers anywhere.
Comparison
How Pantheon stacks up.
| Feature | Pantheon | ChatGPT Pro | Claude Pro | Devin |
|---|---|---|---|---|
| Specialized agent roster (not one model) | Yes | No | No | No |
| Persistent cross-session memory | Yes | No | No | Yes |
| Self-improving skills | Yes | No | No | No |
| Multi-channel (Telegram, Slack, etc.) | Yes | No | No | No |
| Subagent delegation | Yes | No | No | No |
| Cron scheduling | Yes | No | No | No |
| Model-agnostic (200+ models) | Yes | No | No | No |
| Single-tenant deployment | Yes | No | No | No |
| Open-source runtime (no lock-in) | Yes | No | No | No |
| Code execution + terminal | Yes | No | No | Yes |
| Pricing model | Custom enterprise | Per seat | Per seat | Per seat |
Why Prompt Engines
We've been building agent infrastructure for years. Now we run it for you.
We dogfood our own Pantheon daily.
Every agent on your roster runs in our own org first. If it doesn't survive contact with our work — finance, legal, security, growth — it doesn't ship to yours.
Open-source roots.
The Hermes runtime your Pantheon depends on is open source. Not a black box — you can read it, fork it, audit it, and migrate off our infrastructure on any week of any month.
A roster, not a chatbot.
Most "AI for business" is one model behind a prompt. Pantheon is twelve specialized agents with persistent memory, working in concert under a primary orchestrator.
Built by people who run companies, not demos.
The same people building your Pantheon use it to run real businesses. Every archetype was earned in production, not invented for a slide.
FAQ
Common questions.
How long is the engagement?
An initial scoping period covers the first 30–90 days, depending on complexity. After that, the relationship is month-to-month. We earn renewal every cycle.
Who owns the data and the agents?
You do. Your Pantheon's memory, skills, configuration, and history live in your infrastructure (or our racks, on your behalf). Hermes is open source — you can export everything and run it yourself any time.
What if Prompt Engines goes out of business?
You keep the runtime, the configuration, the memory, and the skills. Hermes is open source and well-documented. Migration to self-hosting is a one-day exercise — we provide the runbook at deployment.
Can we use our own models and providers?
Yes. OpenAI, Anthropic, Google, Mistral, Llama, OpenRouter, local Ollama — switchable per agent, per task. Bring your own API keys, or use ours under the cost cap.
What's the cost cap on model and infrastructure?
Set per engagement at scoping, based on your roster size and expected throughput. We absorb the overage. Caps are reviewed quarterly, and you'll see the same dashboards we use to track them.
Can you build agents we don't see in the roster?
Yes — every engagement includes custom agent design. The 12 archetypes are starting points, not a menu. Bring us a role we don't have and we'll build it. Recent additions: Tathagata (Buddhist studies), Muse (creative ideation), Anansi (storytelling).
What integrations do you support?
Anything with an API. Native gateways for Telegram, Slack, Discord, WhatsApp, Email, Signal. MCP servers for tools like Notion, Linear, GitHub, Salesforce, HubSpot, your data warehouse. We add what you need.
How do approvals work for sensitive actions?
The Tirith permission layer enforces four levels: read-only, advise, act-with-approval, autonomous. Configurable per agent and per tool, with inline approvals via your channel of choice. Nothing critical happens without you saying yes.
The Pathway
Book your 90‑minute discovery.
Tell us what you'd hand off first. We'll come back with a scoped pathway in one business day — which agents, which channels, what "live in 30 days" looks like.
- No obligation, no pitch deck
- You leave with a written pathway whether you engage or not
- We sign your NDA before the call if you'd like
- Replied to by a human in < 1 business day