v1.0.0 Released · Open Source

Send a Message.
Get It Done.

AI-powered infrastructure management from your chat. Online or fully offline. No vendor lock-in.

"Trust is earned, not granted."

Star on GitHub Contact Us
Telegram
Slack
Discord
Web App
Teams contact us
Element / Matrix contact us
Features

Built for How You Actually Work

Natural Language

Stop memorizing CLI syntax. Just tell h-cli what you need in plain English.

Online Mode

Connect to the latest AI models via API for maximum reasoning power.

Fully Offline

Run entirely on your hardware with Ollama or vLLM. Your data never leaves your network.

Asimov Firewall

4-layer safety system. Deterministic denylist + independent AI gate. Can't be bypassed.

Teachable Skills

Demonstrate a workflow once, h-cli learns it. Tribal knowledge becomes automation.

Open Source

Every line inspectable. No black boxes, no telemetry. Fork it, modify it, own it.

45 Hardening Items

Non-root containers, read-only filesystems, HMAC-signed messages, network isolation.

Concurrent Tasks

3 parallel tasks per user, cross-user parallelization. Long-running feedback after 30 seconds.

Deployment Profiles

Modular Docker profiles: core, telegram, slack, discord, web, monitor, vectordb, tools.

71% Token Reduction

Plaintext context instead of JSONL replay. Less tokens, same context, lower cost.

CVE Checking

Built-in vulnerability scanning. Identify CVEs across your infrastructure automatically.

Slash Commands

/run, /new, /cancel, /abort, /status, /stats — quick control alongside natural language.

238+
Commits
45
Hardening Items
4 min
CLOS Fabric Deploy
12
Docker Services
Why h-cli

Different Layer, Different Job

Ansible and Nornir are great at running playbooks. NAPALM is great at config management. h-cli does something else entirely.

h

h-cli vs Ansible / Nornir

Input Natural language YAML playbooks / Python
Reasoning AI plans multi-step workflows Predefined task sequence
Learning Learns from every session Static until rewritten
Safety Asimov Firewall (4-layer) --check / dry-run
Interface Telegram, Slack, Discord, Web CLI / AWX

h-cli orchestrates. Ansible/Nornir execute. Different layers — they can work together.

ssh

h-ssh vs NAPALM

Transport SSH, NETCONF, eAPI, telnet NAPALM drivers
Rollback commit confirmed + auto-rollback rollback / compare
Parallelism Built-in parallel execution Sequential by default
AI-aware Designed for LLM tool-calling Not designed for AI
Output Structured for AI reasoning Structured getters

h-ssh is purpose-built for AI-driven automation. NAPALM is great for traditional config management.

Asimov Firewall

Safety Built to Hold

Four hierarchical layers — each with four sub-rules — modeled after Asimov's Laws with TCP/IP-style layering. Lower layers always override upper layers. 45 hardening items across 12 services, enforced by a deterministic denylist and an independent AI judge.

1

Base Laws

  • L1. Protect infrastructure
  • L2. Obey the operator
  • L3. Preserve functionality
  • L4. Respect boundaries
2

Security

  • S1. Never expose secrets
  • S2. Block source access
  • S3. No data exfiltration
  • S4. No privilege escalation
3

Scope

  • R1. Engineering tasks only
  • R2. General knowledge OK
  • R3. No impersonation
  • R4. Infra services only
4

Behavioral

  • B1. Honest about uncertainty
  • B2. Brevity first
  • B3. Graceful failure
  • B4. Explain denials
Intelligence

Two Layers of Knowledge

Built-in vector memory learns from every session. h-knowledge adds expert vendor knowledge on top. Together, they make h-cli smarter with every interaction.

Built-in · Open Source

Every Conversation Becomes Knowledge

Three-tier memory system — from fast session cache to permanent vector search. No external services, no API keys — it all runs locally.

Session Tier

Redis-backed with 24-hour TTL. Fast context for active conversations and ongoing tasks.

Chunk Tier

Permanent disk storage. Every conversation, troubleshooting session, and config change preserved forever.

Vector Tier

Qdrant + MiniLM embeddings. Nightly auto-indexing for semantic search across all sessions.

h-cli memory
> how did we resolve the MTU issue on spine-01 last week?
⟡ Searching vector memory… 3 relevant sessions found
Session #482 (2 days ago) — engineer: halil
spine-01 xe-0/0/0 MTU mismatch: 9192 vs 9216
Fix: set interfaces xe-0/0/0 mtu 9216 + commit confirmed
✓ Resolved — MTU aligned across all spine uplinks
+ ADD EXPERT KNOWLEDGE
h-knowledge API

Add Expert Knowledge on Top

Expert knowledge summarized and rephrased from official vendor docs and RFCs, embedded with a larger model and delivered via REST API.

Built-in MiniLM only

> "Configure EVPN on the leaf switches"

Your conversation history + LLM training data. Knows your past sessions, but vendor details may be generic or outdated.

Good — your context, generic vendor knowledge
With h-knowledge

> "Configure EVPN on the leaf switches"

Your conversations + expert-summarized vendor knowledge and RFC insights, embedded with a larger model. Precise, correct, current.

Best — your context + vendor-accurate, RFC-validated

Vendor Docs

Routing, switching, security, DC fabrics. Summarized from official docs, continuously regenerated.

Standards & RFCs

BGP, EVPN, OSPF, VXLAN — the protocols your network runs on.

API Integrations

NetBox, EVE-NG, Grafana, and more. Pre-vectorized API knowledge.

Custom Packs

We convert your internal wiki and runbooks into queryable vector databases.

Request API Access

Ready to Deploy?

h-cli is open source. Want a customer deployment, live demo, or h-knowledge API access? Let's talk.

Contact Us
Or dive straight in — it's open source on GitHub