Kimchi - 2026 Pricing, Features, Reviews & Alternatives

Kimchi combines a CLI coding agent, a serverless OpenAI-compatible inference API, and an enterprise spend console into a single platform. The coding agent accepts plain-English task descriptions and autonomously reads files, writes code, runs commands, and verifies its own output. Multi-model orchestration classifies each task and routes it to the best-fit model or specialized sub-agent, running security audits, tests, and analysis in isolated parallel contexts. Ferment mode supports structured, plan-upfront project execution with milestone tracking and a single approval gate. Remote persistent sandbox sessions via /teleport continue running after terminal close and reconnect through /sessions or VS Code Remote SSH. Built-in Language Server Protocol support for TypeScript and Go delivers type-aware diagnostics, hover, go-to-definition, references, and rename without additional tooling.

Kimchi Inference provides pay-per-token serverless access to production-ready open-source LLMs through an OpenAI-compatible API hosted at llm.kimchi.dev, supporting models including Claude Sonnet, Opus, Kimi K2, MiniMax M3, and GLM 5. Hybrid model routing directs sensitive or execution-class tasks to self-hosted open-source models and complex reasoning tasks to closed frontier models. The spend console enforces hard budget caps at the per-developer, per-team, per-API-key, and per-org levels in real time, with per-request traces across prompt, model, tokens, latency, and cost. A centralized approved skills registry, MCP-based integrations, persistent memory across sessions, and context window management round out the agent layer. Integrations span GitHub, GitLab, Jira, Confluence, Slack, and Linear.

Enterprise deployments run as a Kubernetes operator inside the customer's own VPC with zero outbound calls and zero telemetry. SSO and RBAC tie to the organization's identity provider. PII filtering strips emails, names, and secrets before prompts reach any model. Secret isolation ensures agents never see raw credentials. SIEM audit logging captures every prompt, completion, and tool call. The platform is SOC 2 Type II certified, GDPR and DORA compliant by design, HIPAA-ready, and FedRAMP Moderate in progress. Migration from Claude Code, Cursor, Continue, VS Code, and Windsurf requires only a base URL change with no code modifications. A free tier is available with no credit card required. The platform targets individual developers, engineering teams, and enterprises, with an enterprise tier positioned for organizations with 50 or more developers and inference spend above approximately $3,000 to $5,000 per month, plus a dedicated Startup Program.