Pick one runtime for kdn. The API notes below update for your choice. You can change this later in settings.
We probe for a local OpenAI-compatible server (typically Ollama on port 11434 or Ramalama). Results update the default-model step.
You can pick a default from the local catalog on the next step.
Choose one cloud provider to configure now (OpenCode can use more later in settings). Enter the credentials for that provider below.
Default models on the next step follow this provider. Add other providers later in Models or workspace settings.
Claude Code uses the Anthropic API unless your team routes traffic elsewhere. Optionally store a key now (encrypted as ANTHROPIC_API_KEY).
agents.jsonThese values map to environment entries for the Claude agent in ~/.kdn/config/agents.json. Run gcloud auth application-default login on the host before starting the workspace.
Mounts (add under claude.mounts in the same file):
No ANTHROPIC_API_KEY is required when using Vertex; credentials come from the mounted gcloud config.
Codex uses the OpenAI API. Optionally store a key now (encrypted as OPENAI_API_KEY unless your agents.json uses another variable).
This is the model OpenCode will use by default in new workspaces. Names match the Models catalog; enable or disable rows there. You can override per session later.
Same rows as Models. Click a line or use the Use control to pick your default for local models.
| Status | Name | Size | Runtime | Use |
|---|---|---|---|---|
|
qwen3-code
GGUF · loaded in memory · OpenCode default
RAM usage: 5.8 GiB
|
5.4 GB | Ollama | ||
|
llama3.2:3b
GGUF · loaded in memory
RAM usage: 3.93 GiB
|
2.0 GB | Ollama | ||
|
mistral:latest
GGUF · loaded in memory
RAM usage: 4.4 GiB
|
4.1 GB | Ollama | ||
|
qwen2.5:7b
Not loaded — Ramalama stopped
RAM usage: N/A
|
4.7 GB | Ramalama |
| Status | Name | Size | Age | Use |
|---|---|---|---|---|
|
ibm-granite-3.3-8b-instruct
OpenShift AI · Apache-2.0
VRAM est. 8.2 GiB
|
4.8 GB | 2 weeks | ||
|
mistral-small-internal
OpenShift AI
VRAM est. 11 GiB
|
6.1 GB | 1 month | ||
|
llama-3.1-70b-instruct
OpenShift AI · pending approval
RAM usage: N/A
|
— | — |
Rows match the cloud provider from step 1. Align names with the Models catalog after onboarding.
| Status | Name | Size | Provider | Use |
|---|---|---|---|---|
|
gpt-5.4-medium
OpenAI · general reasoning
|
— | OpenAI | ||
|
gemini-2.5-flash
Google · low latency
|
— | Gemini | ||
|
claude-sonnet-4-5
Anthropic API · via OpenCode
|
— | Anthropic | ||
|
gpt-5.2
Azure OpenAI · map to your deployment name in OpenCode
|
— | Azure | ||
|
openai-compatible
Your base URL · use the model id your server exposes
|
— | Custom |
Cloud catalog on Models — Anthropic rows (enabled). Click a row to choose.
| Status | Name | Size | Runtime | Use |
|---|---|---|---|---|
| — | Anthropic | |||
| — | Anthropic |
Same cloud model IDs; routing via Vertex in step 1.
| Status | Name | Size | Runtime | Use |
|---|---|---|---|---|
| — | Anthropic | |||
| — | Anthropic |
Composer and OpenAI rows (enabled), same order as Models → Cloud.
| Status | Name | Size | Runtime | Use |
|---|---|---|---|---|
| — | Composer | |||
| — | OpenAI | |||
| — | OpenAI |
This step looks for MCP tool configs and agent skills under your project roots. After detection, matches are registered in Kaiden for your agents. Knowledge bases are configured elsewhere.
Kaiden walks your project roots and looks for MCP tool manifests under .mcp/ and reusable agent skills. Matches are imported into Kaiden as soon as the scan finishes—no pick list. Knowledge bases and other catalogs are not part of this step.
Everything listed below is registered in Kaiden automatically: MCP servers are available to agents, and skills show up in the catalog. You can review or remove items later under AI Assets.
Your Kaiden environment is configured. Filesystem and network for the sandbox are chosen when you create a workspace. Then run kdn workspace start in your terminal to work with the coding agent.