Axon deploys a complete AI coding platform inside your secure network. IDE autocomplete, web chat, code review agents, and autonomous tooling — powered by best-in-class open-source models on your own GPU hardware. Nothing ever leaves your perimeter.
The full stack of modern AI coding tools — no byte of your code ever touches an external server.
Real-time inline tab-completion and a sidebar AI chat panel wired directly to your Axon server. Works in VS Code, the full JetBrains suite, and Android Studio — installed offline via packaged plugins.
Browser-based multi-user chat. Upload docs, query internal codebases with RAG-grounded answers, share workspaces across teams. No install required.
Terminal-first coding agent for multi-step, multi-file agentic tasks. Plan, act, test, and commit — scripted or interactive, with CI/CD integration.
SAST plus LLM reasoning in one pipeline. Inline MR comments, memory-safety scanning, offline CVE dependency audit, and configurable CI gate blocking.
Scheduled and event-triggered agent VMs — each with its own RBAC scope and least-privilege MCP tool access. Managed from the Axon Console.
Full governance portal. RBAC, department management, LDAP/AD integration, usage budgets, metering, alerting, model placement, agent registry — everything in one place.
Via IDE plugin, browser, or CLI — over internal LAN through Axon Gateway on TCP/443. TLS-terminated, JWT-validated, zero external hops.
Axon Gateway dispatches to Abot Studio, the Inference Engine, Code Review Agent, or MCP pool — based on path and JWT claims.
CUDA-accelerated token generation on your hardware. OpenAI-compatible API, streaming SSE, KV-cache, batching — no external calls, no telemetry.
Tokens streamed back to the developer. Usage metered, no prompt content stored. Audit event written. Air-gap verified — nothing crosses the perimeter.
Every component runs on your hardware, on your network. One ingress point. No outbound traffic. Verified at every deployment.
VS Code · JetBrains · Android Studio
Autocomplete · Chat · Analysis
Multi-step agentic tasks
MCP tool client · CI/CD
Axon Abot Studio
Axon Console (admins)
Encrypted NVMe · GPG-signed
SHA-256 manifest · SBOM
Single TCP/443 ingress
GPU PCI passthrough
OpenAI-compat API
Streaming · KV-cache
Multi-user web chat
RAG · pgvector
Doc workspaces
RBAC · LDAP/AD
Model placement
Metering · Audit
Scheduled · event-triggered
Least-privilege · own RBAC
Git · FS · CI · Code-search
Docs · Ticketing MCPs
GitLab · Gitea
MR webhooks ↔ Git MCP
Inline comments
Jenkins · GitLab CI
Build logs ↔ CI MCP
Pass / BLOCK gate
Bind → JWT mint
Group → dept mapping
Auto-provision
Splunk · Elastic
QRadar · ArcSight
Syslog/TLS · CEF
Axon ships with a curated set of open-source coding and reasoning models, all deployed offline on your hardware. Prefer a custom or fine-tuned model? Axon supports that too.
Highest-capability coding model for IDE chat, agentic multi-step tasks, and complex code generation with deep context understanding.
Optimised for real-time inline tab-completion. Low VRAM footprint, sub-second first-token latency under concurrent IDE load across your whole team.
General-purpose assistant model for documentation chat, architecture Q&A, and non-code reasoning queries via the Abot Studio web interface.
Have a fine-tuned or domain-specific model? Axon's inference engine is model-agnostic. Drop in any GGUF or safetensors model and it works immediately with no reconfiguration.
No default route to the internet. Verified post-install by the built-in Air-Gap Verifier — port sweep, outbound DNS/TCP/HTTPS confirmation, CIS L1 re-scan.
All traffic enters on TCP/443 only via the Axon Gateway. TLS 1.3, AEAD ciphers, HSTS, OCSP. LAN-only ACL — no public firewall rules required.
Every component ships with telemetry, analytics, and call-home behaviour fully disabled. Verified at the network layer. No prompt content stored by default.
LUKS full-disk encryption. Model weights, conversation history, documents, and logs reside only on your server. Zero external replication.
GPG-signed delivery bundle, SHA-256 manifest, SBOM in SPDX and CycloneDX formats, digest-pinned container images. Every component fully traceable.
JWT auth, LDAP/AD pass-through, RBAC per department, session revocation on deactivation. API keys stored in OS keychain — no plaintext credentials.
Six clearly defined phases with formal validation gates between each. Nothing moves forward until the previous layer is proven, giving your security team full visibility throughout.
Kickoff, environment survey, workstation audit, offline bundle preparation, SBOM & licence review. Scope and security posture signed off before any installation.
Server racking, OS hardening, CUDA configuration, firewall rules, and air-gap validation. Security officer sign-off required before software deployment begins.
Inference engine, Abot Studio, Axon Console, and Gateway deployed. Models loaded, benchmarked, and demonstrated to the customer before integration begins.
IDE plugins installed across all developer workstations via automated scripts. Verified per IDE type including Android Studio, with pilot group feedback before full rollout.
Pilot developers use Axon in real daily work. Formal acceptance test checklist executed, performance validated, air-gap re-verified, security officer sign-off obtained.
Developer and admin training delivered on-site. Full documentation and runbooks handed over. Active hypercare support period begins with regular check-ins post go-live.
Each phase concludes with a formal validation gate requiring written sign-off from both vendor and customer leads. Delivery timelines are scoped during Phase 1 based on your environment size and requirements.
Our team handles hardware specification, offline model delivery, full configuration, integration, and training — end-to-end, inside your perimeter.