Lobstack is a managed cloud platform for deploying autonomous AI agents. Each agent runs on a dedicated virtual machine with persistent memory, 117+ integrations, and goes live in under 90 seconds with zero DevOps.

How long does it take to deploy an AI agent?

You can deploy an AI agent on Lobstack in under 90 seconds. Describe what your agent should do, pick a model, and Lobstack handles all the infrastructure — no Docker, Kubernetes, or server setup needed.

What AI models does Lobstack support?

Lobstack supports 24 AI models including Claude Opus and Sonnet from Anthropic, GPT-5 and GPT-4o from OpenAI, Gemini 2.5 Pro from Google, plus DeepSeek, Mistral, and Grok. You can hot-swap models without losing agent memory or configuration.

Is Lobstack secure for enterprise use?

Yes. Every agent runs in a gVisor-sandboxed VM with AES-256 encryption at rest, TLS 1.3 in transit, HashiCorp Vault for secrets, row-level security on all tables, and full audit logging. Lobstack is SOC 2 aligned.

Docs/Agentic Enterprise/Overseer + Operator

Overseer · Operator

Agentic Enterprise

Beyond single agents — a self-managing system. The Operator lets each agent build and run a business function on its own; the Overseer is the fleet brain that learns across every agent and proposes improvements you approve. Together they form a loop that observes, plans, executes, measures, and improves.

✅

The loop

Set objective → auto-plan → execute steps with tools → report & track KPIs → Overseer watches, proposes, applies, and measures → the fleet learns. No human step required to keep it running — only to approve fleet-wide changes.

Operator — agents that operate a business#

Give an agent an objective (optionally a KPI metric, a target, and a spend budget). The Operator turns it into an executable plan and works it autonomously:

Plan — an LLM planner (any model via the Gateway) decomposes the objective into 3–6 concrete, ordered steps.
Execute — an hourly loop kicks the agent to work the next step using its 150+ tools and integrations (search, comms, CRM, code, payments).
Report — the agent marks each step done/blocked with a result and logs KPIs.
Track — KPIs auto-refresh from live platform data; goal progress updates against the target.
Budget — payments tied to the goal draw from its budget and never exceed it.

Operator tools (the agent uses these itself)#

Tool	What it does
operator_set_goal	Define an objective + optional metric / target / budget.
operator_add_step	Add an ordered, executable step to a goal's plan.
operator_update_step	Mark a step in_progress / done / blocked with a result.
operator_log_kpi	Record a KPI measurement; updates the linked goal's progress.
operator_status	Read current goals, plan steps, and recent KPIs to decide what's next.

You can also drive it from the dashboard Operator tab — add an objective, set a budget, generate a plan, and watch steps and KPIs progress.

API#

/api/agent/operator

# set a goal with a metric, target, and spend budget
POST /api/agent/operator  { "action":"set_goal", "objective":"Grow MRR to $10k",
                            "metric":"mrr_usd", "target_value":10000, "budget_usd":500 }
# auto-generate a plan for it
POST /api/agent/operator  { "action":"plan", "goalId":"…" }
# read current operating state
GET  /api/agent/operator   # → { goals:[{…, steps:[…], spent_usd}], kpis:[…] }

Overseer — the fleet brain#

The Overseer is a meta-intelligence over your entire agent fleet. It runs on a schedule and:

Stage	What happens
Observe	Aggregates privacy-safe metrics across every agent — cost, error rate, latency, skill usage, payments. Never reads prompts or message content.
Learn	A deterministic rule engine plus an optional LLM analyst (any model via the Gateway) find cost / reliability / skills / payments / growth opportunities, and cross-fleet lessons.
Propose	Each insight can carry a suggested action (switch to managed routing, tighten a spend policy, promote a self-optimized prompt, or advise). Nothing auto-applies.
Approve	An admin approves or dismisses each action in the admin panel — the human stays in control.
Apply	Approved changes are pushed live to the agent's VM (hot-reload), and recorded.
Measure	After a learning window the Overseer measures the real effect and folds it into fleet-lesson confidence — so it gets better at its own decisions.

⚠️

Enterprise isolation

Cross-fleet learning uses privacy-safe aggregates only — numbers and labels, never raw content. The Overseer is org/tenant-scoped: per-org rollups and an admin org filter keep each enterprise's data separate.

Overseer ↔ Operator feedback#

The Overseer reads Operator goals and surfaces the ones at risk — behind pace against a due date, or stalled with no progress — so business outcomes, not just infrastructure metrics, drive its recommendations.

Powered by Lobstack Pay#

When an Operator step needs to spend — pay an API, a vendor, or another agent — it uses Lobstack Pay (LSP-1): USDC on Base (settled on-chain by the LobstackSettlement contract, gas sponsored) or instant USD-pegged Credits, always within the wallet policy and the goal's budget.

✅

Status

The Operator (planner, execution loop, KPIs, budgets) and the Overseer (rules, outcome learning, org scoping, approve/apply) are live. The LLM analyst and planner activate once a managed model key is configured for the Gateway; on-chain USDC settlement activates once the chain env is set. Each degrades gracefully until then.

Templates Library Lobstack Pay (LSP-1)