It reads. It never touches.
Auspex is an on-premise, read-only fault-diagnosis assistant for your network. It runs entirely inside your closed environment (air-gapped, on your own hardware), powered by Kestrel, a compact model trained in our synthetic network lab.
V1: Arista EOS + Linux · Turkish diagnosis output · Fixes require human review
The networks that need help most can't phone home.
Enterprises won't expose their network configuration to an outside service, and many of the sites that need diagnosis most have no internet at all. A cloud AI assistant simply cannot run there.
Auspex answers that constraint directly. Everything (the assistant, its tools, and the model) lives on the customer's own hardware, inside the closed network. Nothing leaves the site.
A data factory, not a data grab.
Auspex learns from faults we manufacture in a virtual lab, so it never needs a single byte of your production network to get good.
Synthetic lab
A virtual lab of Arista EOS switches and Linux hosts. We create fault scenarios without collecting customer network data.
Expert distillation
A large teacher model diagnoses each fault using only read-only tools, then a deterministic verifier keeps solely the traces that are provably correct.
The model learns
The compact Kestrel model learns from those verified traces, distilling an expert's judgment into something small enough to run offline.
Kestrel: small on purpose.
Kestrel v1 is a Qwen3-8B model fine-tuned for network fault diagnosis, distributed as a roughly 5 GB Q4_K_M GGUF file.
It investigates with tools and proposes a diagnosis. V1 outputs Turkish and supports Arista EOS and Linux. An engineer reviews the diagnosis and suggested fixes.
Kestrel is scored against its expert teacher on a held-out exam before any performance claim is published.
Compact
Small enough to run on a single modest on-site machine, no GPU cluster, no data center.
Offline
After installation, inference runs locally. Disable the CLI version check with AUSPEX_NO_UPDATE_CHECK=1 for offline use.
Purpose-built
Trained on one job (network fault diagnosis), not a general chatbot stretched to fit.
Read-only hands
Reasons over the same read-only tools as its teacher. It can look; it can never change a thing.
Diagnosis without changing device configurations.
Auspex collects evidence with permitted commands and leaves configuration changes to the engineer.
Read-only by architecture
The model can never change a configuration. Read-only is enforced at the transport layer (a three-layer, fail-closed allowlist), not by a prompt that can be talked around.
Air-gapped by design
Auspex runs 100% inside your closed network. No telemetry, and no data leaves the site. The only outside call is the CLI's version check, and AUSPEX_NO_UPDATE_CHECK=1 turns it off.
Updated on your terms
Model downloads are checked with SHA-256. Offline transfer and digitally signed deployment packages are separate from the current installer.
What Auspex diagnoses today.
V1 supports Arista EOS switches and Linux hosts. Cisco, Juniper and FortiGate support is not included in this release.
We graded the model. Here is the number.
Recorded results on 150 held-out lab scenarios across 51 unseen topologies, using read-only diagnostic tools and a deterministic verifier. Scenarios include faults, healthy networks and insufficient information.
Partial credit (right class, incomplete detail) is counted as incorrect in the headline number. · Kestrel v1 · 2026-07-23
Methodology
- 150 questions across 51 network topologies the model never saw in training
- Scenarios run in a virtual network and include injected faults, healthy networks and insufficient information
- The tools are shared; base Qwen used a 16,384-token context and Kestrel used 24,576 tokens
- Deterministic verifier scores fault class, culprit device and root cause, no human judging, no LLM judging
These are controlled lab results, not measured production accuracy. The Claude Sonnet reference includes six infrastructure errors in the 150-question denominator. Context settings differ between base Qwen and Kestrel.
Install Auspex. Download Kestrel.
Install and start Ollama first, then install the CLI and download Kestrel. The first download requires internet access.
Auspex is the diagnostic application; Kestrel is the model it runs through Ollama. The CLI installer downloads Auspex from GitHub. Then auspex setup downloads and registers Kestrel.
- ~5 GB model file; ~9 GB resident memory at the default context; speed varies by hardware
- V1: Arista EOS + Linux · Turkish diagnosis output · Fixes require human review
- Results cover a virtual lab. Multi-fault, static-route and some ACL scenarios remain challenging.
- CLI and weights: Apache-2.0
curl -fsSL https://github.com/Calexis-Labs/auspex-cli/releases/latest/download/install.sh | shirm https://github.com/Calexis-Labs/auspex-cli/releases/latest/download/install.ps1 | iexauspex setupauspex doctorauspex chat --demoBring Auspex to your network.
Questions, feedback, or help deploying Auspex inside your network? Get in touch.