Enterprise AI Infrastructure | Kimchi Your team cooks. You are the Chef Get visibility and control over your AI tools usage and spending. 1-minute to set up. Talk to us →Book a demo DEPLOYMENT TOPOLOGYRESIDENT YOUR VPC · DATA STAYS INSIDE DEVELOPERS · IDEs · AGENTS Your team, your tools cursorvscodekimchi cliclaude code CONTROL PLANE Kimchi orchestrator routing · policy · audit DATA LAYER Your databases RAG · context · memory GPU CLUSTER · YOUR ACCOUNT Open-source models on your hardware minimax-2.7glm-5kimi-k2nemotronyour-finetune.gguf All inference loops inside your perimeter External calls · openai.com · anthropic.com● blocked by policy Data sovereignty Your cloud. Your data. Your rules Open-source models run inside your cloud account on AWS, GCP, or Azure. Prompts, completions, and code never leave your Cloud. OTHER AI VENDORS Your code, their cloud. Closed APIs send every prompt, every file, every secret to data centers you don't own and to jurisdictions you may not be allowed to ship to. your code→their server→model ↳ residency unknown · subject to CLOUD Act · audit gaps KIMCHI ENTERPRISE Your code, your perimeter. Every model, primary and fallback, runs inside your VPC. Air-gappable end to end. Nothing phones home. You decide what crosses the boundary; the default is nothing. your code→your vpc↺your model ↳ residency · yours  ·  CLOUD Act · n/a  ·  full audit trail Read more about security → Security & Compliance Built for regulated teams. Kimchi Enterprise adds identity, audit, and compliance controls on top of a VPC-native architecture. Your data stays inside your perimeter, and you have the proof points to back it up. ID Identity & access SSO/SAML/OIDC, group-based RBAC, and SCIM-ready user provisioning. Tie agent policies to your IdP groups. AU Audit & logging Per-request traces for every prompt, completion, and tool call. Ship logs to your SIEM and retain an immutable audit trail. SC Secrets isolation Credentials stay in your Vault or secrets manager. Agents request scoped secrets at runtime and never see raw keys. DT Data handling No training on your data. No prompt or code stored on Kimchi-operated infrastructure in Serverless or Self-Hosted modes. CM Compliance posture SOC 2 Type II, GDPR-ready, and HIPAA-ready architecture. BAA available on request. BY BYOK frontier Need Claude, GPT-5, or Gemini? Route to your own provider keys with a clear, explicit data flow and no Kimchi interception. Read the full security overview → Governance & guardrails No rogue merges. No leaked secrets.  No runaway spend. Every coding agent in your org routes through the Kimchi Proxy. Budgets, PII filtering, usage metrics, and an approved skills registry are enforced before requests hit a model and surfaced in the Kimchi web app. Coding agents Kimchi Coding Claude Code Cursor Codex ↑↓ ↑↓ ↑↓ ↑↓ Kimchi Proxy Policy & enforcement layer Budgeting Hard caps per team, dev, or API key with graceful downgrade at threshold. PII filtering Strip emails, names, and secrets before prompts ever hit a model. Usage metrics Per-request traces across prompt, model, tokens, latency, and cost. Skills registry Centralized catalogue of approved tools, prompts, and MCP skills. ↑↓ Kimchi Web App Reports & insights ▤ Analytics Report Total requests, token consumption, spend trend, and breakdown per team. ↗ Adoption Report Seat utilization, active users over time, onboarding progress. <> Coding Agent Report Sessions, completions, acceptance rate, estimated time saved. $ Cost & Billing Usage per model, free credit consumption, billable spend this period. Cost visibility See exactly who, what, and where your AI spend is going. Per-developer, per-team, per-model, per-tag in real time. Forecast spend before scaling. Stop discovering the invoice on the 1st. app.kimchi.dev / enterprise / cost-insights ● LIVE Cost insights · $12,408 this month · 312 developers FORECAST · EOM $18.4k ▼ 71% vs claude SESSIONS 142,891 ▲ 28% wow AVG / DEV / MO $39.80 vs $200+ claude GPU UTIL 62% 28 H100 · 12 hibernated DEPARTMENT DEVS SESSIONS SPEND BUDGET · USED Platform Engineering 94 52,401 $4,820 64% Product Engineering 142 68,124 $5,310 88% Data Science 38 14,892 $1,420 35% SRE / Infra 38 7,474 $858 43% Hybrid model routing Stop paying architect rates for every task. Use a powerful model for reasoning, route execution to cheaper self-hosted open-source models. Hybrid mode keeps the best of both worlds, and the routing decision is automatic. KIMCHI AUTO Model Orchestration Routes each request based on task complexity, sensitivity, and team policy. CLOUD · COMPLEX REASONING ~12% of requests Closed models · best-in-class plans Architecture decisions, complex debugging, long-context analysis, routed to Claude / GPT-5 when the team allows external calls. claude sonnetgpt-5gemini-2.5 SELF-HOSTED · EXECUTION ~88% of requests Open-source · in your VPC Code generation, refactors, tests, and embeddings run entirely on your GPUs at a fraction of the cost. Default for sensitive tasks. minimax-2.7glm-5-fp8nemotronyour-finetune.gguf Real blended workload, last 90 days · 40–70% cost reduction without sacrificing output quality on complex tasks −62% Zero-friction migration One command. From your existing AI tools. kimchi setup auto-detects Claude Code, Cursor, Continue, VS Code, Windsurf and migrates the endpoints automatically. OpenAI-compatible API. No code changes required. Start on Kimchi serverless, graduate to self-hosted whenever compliance or cost demands it. Auto-detects every coding tool already installed Same SDK, same workflow. Only the base URL changes Migrate MCP servers, skills, and config in one prompt Graduate to self-hosted with a single config flag kimchi setup · enterprise migration $ kimchi setup --enterprise scanning installed tools… ✓ Claude Code · v2.4.1 · 47 configs · 12 skills ✓ Cursor · v0.42 · OpenAI base URL set ✓ VS Code · Continue · v1.0 · 3 model configs ✓ Windsurf · v0.6 · cascade mode active migrating MCP servers · github, postgres, filesystem… remapping endpoints → kimchi.your-corp.io/v1 ✓ migration complete · 4 tools · 0 changes required to your code $ kimchi opening kimchi · connected to kimchi.your-corp.io Kimchi Harness · Enterprise Out-of-the-box connectors for the tools  your engineers already use. MCP-based integrations, context window management, persistent memory across sessions, spec-driven development workflows. A full-stack coding platform that never leaves your perimeter. GitHub Enterprise PRs, diffs, comments, status checks. Read-only or scoped writes. GitLab Full MR + pipeline integration. Self-managed instances supported. Jira Read tickets, link commits, create issues from PR findings. Confluence Spec docs, runbooks, and ADRs accessible during planning phases. Slack Notifications, DMs, channel triggers. Per-team routing rules. Linear Issue graphs, cycles, projects. Bidirectional with PRs. Postgres / MySQL Scoped queries against your DB for RAG and analysis. S3 / GCS / Azure Document and artifact storage. Signed URLs handled internally. Okta · SAML / OIDC SSO and RBAC tied to your IdP. Group-based agent policies. Vault / Secrets Mgr Credential isolation. Agents never see raw secrets. Datadog · Splunk Log every prompt, completion, tool call to your SIEM. Custom MCP Any HTTP endpoint becomes a typed tool. @tool decorator. Why self-hosted? The questions enterprises ask. Is self-hosted really cheaper at scale? Yes, once you cross ~$3-5k/month in inference spend. We model your break-even in the first call; most enterprises with 50+ developers are well past it. Below that, Kimchi Serverless is cheaper. How does quality compare to closed models? For execution-class tasks (code generation, refactors, tests, embeddings), open-source models match or exceed Sonnet on real workloads. For planning and complex reasoning, hybrid routing keeps closed models in the loop when you allow it. Which compliance frameworks are you ready for? SOC 2 Type II today. GDPR and DORA by design. HIPAA-ready architecture (BAA on request). FedRAMP Moderate in progress. Customer-specific audits supported. What's the operational overhead? Kimchi runs as a Kubernetes operator inside your cluster. Autoscale, hibernation, and monitoring are managed. Your team manages identity, network, and the underlying nodes. Most customers spend <1 SRE-day/month on ops. Are Open Source AI Models Actually Secure? Every model, including fallbacks, runs inside your VPC with zero outbound calls, zero telemetry, and no updates without your approval, deployed from a single signed bundle. How does migration from serverless work? One config flag. Change base_url from api.kimchi.dev to kimchi.your-corp.io. Same API, same SDK, same code. Most teams flip the switch in under an hour. Ready to bring AI coding inside your perimeter? Talk to our enterprise team. We'll model your break-even, walk you through the architecture, and have you on Kimchi in your VPC in weeks, not quarters. Talk to us →Book a demo Custom rate limits Dedicated CSM SLA · SSO · audit logs Procurement-friendly contracts