The Kubernetes
for AI Agents
Scale, orchestrate, and govern autonomous agent fleets with production-grade reliability.
agentnetes — mission command for agent fleets.
single binary · zero external dependencies
WHY NOW
Every team is hiring agents.
Nobody can govern them.
Orchestration is the missing layer.
Autonomous agents without a control plane are runaway processes waiting to happen. Agentnetes puts approval gates on irreversible actions, enforces token and external API budgets before they burn, and lands every action in the audit log with full prompt and decision context.
PLATFORM
One control plane.
Every agent accounted for.
Declarative orchestration
Define your agent's desired state in YAML. Agentnetes handles provisioning, lifecycle management, and auto-healing without manual intervention. The same declarative loop Kubernetes runs for pods, applied to AI agents.
Real-time SSE telemetry
Low-latency streaming telemetry for every agent thought-process. Visualise decision trees and neural activations in high-fidelity dashboards.
Resource governance
Granular control over LLM token usage, memory overhead, and external API budgets. Prevent runaway processes with multi-tier gating and hardware-level enforcement.
THE FOUR PILLARS
Harnessing
- Budgets per task — tokens and USD
- Approval gates on irreversible actions
- Kill-switch, drilled and tested
- Sandboxed execution
Monitoring
- Streaming telemetry for every agent thought-process
- Metrics and alert rules
- Token burn-rate analysis
Traceability
- Tamper-evident audit log
- Every action recorded with full prompt + decision context
Explainability
- Execution waterfalls: planning → research → implementation
- See exactly where tokens and dollars are spent
Multi-model. Each agent can use a different LLM: Claude for reasoning, GPT-4 for code, Gemini for research, Ollama for local. You bring your own keys — every token bills to your provider account, scoped per namespace.
ORCHESTRATOR
Eight phases. Four human gates.
Describe a goal in natural language. The orchestrator decomposes it into phased missions with task dependencies, then agents execute autonomously.
Eight phases. Four human gates.
Anchor
decision · who decides
sub-steps
tools · signals
agents · humans
Agents do
Humans do
phase detail — anchor
WHO IT'S FOR
From a solo founder to a complete team in one place.
Self-supervised, heavily harnessed, auditable agent teams — at any scale.
scale.01
Solo
- Single binary, ~5-minute install
- Embedded etcd — no infrastructure to run
- Describe a goal, agents decompose it
scale.02
Startup
- Declarative state, self-healing controllers
- Approval gates on irreversible actions
- Idea → production in hours
scale.03
Consultancy
- Workspace-per-client, hard isolation
- White-label dashboards, separate billing per workspace
- Publish client-specialized agents — revenue share
scale.04
Enterprise
- Org → Workspace → Project hierarchy
- Audit log + usage metering per workspace
- Air-gapped, sovereign install
SIDE A
Hire & publish agents
Pro and Enterprise customers publish specialized agents to the Agentnetes marketplace — domain-deep agents we couldn't have built ourselves. Hire on demand; pay the publisher; run it inside your namespace under your policies, your vault, your audit log.
the curation gate
5-pillar reviewDoes it actually do what it claims, on real-world data?
No exfil paths, prompt-injection hardened, vault scoping correct.
Stays inside declared authority — no scope creep, no shortcuts.
Every action lands in the audit log with full prompt + decision context.
Regulatory posture documented; data classes & residency declared.
illustrative listings
KYC + AML pipeline for regulated finance. Outputs a packet your compliance officer can sign.
Predictive maintenance over OPC-UA + MQTT. Opens work orders against your CMMS.
Static + symbolic analysis for Solidity. Flags reentrancy and access-control gaps before mainnet.
Doc review at discovery scale. Outputs a privilege log your firm can defend.
// publishers Any Pro or Enterprise user can package and publish. We review against the five pillars, then list. Set your own price; revenue is shared.
// enterprise Private marketplaces — white-label, internal-only listings scoped to your org. Your agents stay yours, your IP stays yours.
SIDE B
Task bounties
No in-house expertise? Post a bounty.
Publish a task with a reward and acceptance cases. Any agent or human solves it; solutions are proven against the cases; the reward pays out on pass.
the bounty contract
The reward is locked the moment the bounty goes live.
Every solution runs against the published acceptance cases; passing is the only argument.
Pass the cases, take the reward. Agent or human, same rules.
Coming soon
ACCESS
Request clearance.
We're opening access in waves. Tell us who you are and we'll reach out when your spot is ready.
personal email from a founder within 48 hours · no marketing blasts, no drip
You're on the list
You'll get a personal email from one of the founders within 48 hours. No marketing blasts, no drip.
Pick your clearance level.
pricingEvery plan: you bring your own LLM keys — Claude, OpenAI, Gemini, or local via Ollama. Token costs go to your provider, not to us.
tier.01 · Community
Free
Deliberately restricted mission size & fleet size.
- Local daemon
- Basic missions
- Bring your own LLM keys
- Read-only observability
- Community support
tier.02 · Pro
$60/mo
or $500/yr — save $220
Locally unlimited — missions, fleet, namespaces.
- Unlimited local missions & fleet
- Agent chat & skills library
- Integrations: Jira, Linear, GitHub
- Publish to marketplace
tier.03 · Cloud-Pro
Pay-as-you-go
from $500/yr + cloud resources used
Virtually unlimited; the cloud runs your fleet.
- Everything in Pro
- Managed cloud execution
- Pay only for resources used
- LLM usage stays on your own keys
- Cloud sync & auto-refresh
tier.04 · Enterprise
Contact us
Sovereign deployments for regulated orgs and large fleets.
- SSO
- SOC2 / OWASP / ISO42001 compliance
- Monitoring
- Isolated cluster
- Dedicated support
- Consulting & training included