Kode reads a set of environment variables at startup to configure its LLM backend, resolve file paths, and control interactive behavior. You can set these in your shell profile, aDocumentation Index
Fetch the complete documentation index at: https://kode-f177b001.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
.env file at your project root, or in your CI/CD pipeline. Where a variable overlaps with a field in .kode/kode.json, the environment variable takes precedence for API keys and endpoint overrides.
LLM & API keys
The primary API key for the configured LLM provider. This takes precedence
over all other key variables (
OPENAI_API_KEY, KODE_PRO_API_KEY). Set this
when using the Kode Gateway or any provider that accepts a Bearer token via a
single key.Standard OpenAI API key. Kode uses this as a fallback when
KODE_LLM_API_KEY
is not set and no other key is provided. If you are already exporting this
variable for other tools, Kode picks it up automatically.Kode Pro API key. When this is set and no other key or endpoint overrides are
present, Kode bypasses the public gateway and routes directly to
api.deepseek.com using the deepseek-v4-flash model. This is the fastest
zero-config path for Pro subscribers.The LLM API endpoint Kode sends chat completion requests to. Override this to
point at a local model server, a corporate proxy, or a self-hosted inference
endpoint. The value must be a base URL; Kode appends
/chat/completions
automatically.Override the model Kode uses, bypassing the
model field in kode.json. Use
the same provider/model format as the config file, or a bare model name when
pointing at a local endpoint.TUI & paths
Custom path for the TUI bundle directory. On first run, Kode downloads the TUI
bundle (~52 MB) from GitHub Releases and extracts it to
~/.kode/tui. Set
this variable to use a different location — useful when ~ is on a slow
filesystem or when you want to share a pre-extracted bundle across users.Custom URL to download the TUI bundle archive from. When set, Kode fetches the
tui-bundle.tar.gz from this URL instead of GitHub Releases. Use this to
point at an internal mirror or a pinned artifact in air-gapped environments.Override the path Kode uses for the
.kode config directory. By default Kode
looks for .kode/kode.json in the current working directory. Set KODE_DIR
to use a directory at a different absolute path — for example, a shared config
location on a build server.Set to
1 to disable all automatic install prompts. When Kode detects that a
dependency (Bun, the TUI bundle, or Sicario) is missing and this variable is
set, it fails immediately instead of asking to install. Always set this in CI
to prevent pipelines from hanging on interactive prompts.Override the path to the Sicario SAST binary. Kode normally installs Sicario
into your
.kode/ directory via kode init or kode install sicario. Set
SICARIO_PATH to point at a pre-installed binary — useful when distributing
a company-wide Sicario installation or when the binary lives outside the
project tree.Server authentication
Password for the Kode server. Required when running
kode serve to protect
the HTTP API from unauthorized access. Clients must supply this password as
a credential when connecting.Username for the Kode server. Paired with
KODE_SERVER_PASSWORD to form the
basic authentication credential for the server. Defaults to kode if not
set.Usage notes
Environment variables take precedence over
.kode/kode.json settings for API
keys and endpoints. If both KODE_LLM_ENDPOINT and providers.gateway_url
are set, the environment variable wins. This lets you override per-project
config at the shell level without editing committed files.Zero-config mode
If none ofKODE_LLM_API_KEY, KODE_LLM_ENDPOINT, OPENAI_API_KEY, or KODE_LLM_MODEL are set, Kode enters zero-config mode and routes to the public Kode Gateway at https://api.trykode.xyz/v1 using deepseek-v4-flash. The public gateway is rate-limited to 20 requests per day per IP on the free (Lite) tier.
Related
kode.json reference
Full reference for all fields in the project config file.
Providers
Configure AI providers, API keys, and model strings.