config.toml Reference

Complete reference for ~/.opscode/config.toml — model defaults, display settings, tool permissions, and interpreter configuration.

OpsCode reads its main configuration from ~/.opscode/config.toml. Use /config in an interactive session to view and modify settings, or opscode config show from the shell. See Configuration for how settings resolve across environment variables, config.toml, and defaults.

---

File location

~/.opscode/config.toml

Create this file manually or let OpsCode create it automatically when you first change a setting via /config or /model.

---

Models

[model]
default = "anthropic:claude-opus-4-7"     # Default model (provider:model-name format)
reasoning_effort = "medium"               # low, medium, or high

Key	Type	Default	Description
default	string	(none)	Default model specifier in provider:model-name format
reasoning_effort	choice	"medium"	Reasoning effort level for supported models: low, medium, high

Set the default model from the CLI:

opscode --default-model anthropic:claude-opus-4-7

Or use /model inside an interactive session to switch models on the fly.

---

Providers

Configure provider-specific settings and endpoints:

[providers.openai]
enabled = true
base_url = "https://api.openai.com/v1"
models = ["gpt-4.1", "gpt-4.1-mini", "o3-mini"]

[providers.anthropic]
enabled = true

[providers.ollama]
enabled = true
base_url = "http://localhost:11434"
models = ["llama3.3", "qwen2.5-coder"]

Key	Type	Description
enabled	bool	Whether this provider is active
models	list[string]	Model identifiers available from this provider
base_url	string	Override the default API endpoint
base_url_env	string	Env var name for the base URL override
api_key_env	string	Env var name for the API key
class_path	string	Custom LangChain model class path
params	table	Extra kwargs passed to the model constructor
display_name	string	Human-readable provider name

Endpoints, keys, and gateways

If you use an API gateway, proxy, or private VPC endpoint, configure base_url:

[providers.openai]
base_url = "https://my-gateway.corp.internal/openai/v1"

Or use /auth inside a session to configure keys and base URLs interactively.

---

Display / UI

[ui]
theme = "dark"                    # Color theme
show_scrollbar = false            # Vertical scrollbar in chat area
show_timestamps = true            # Timestamps on messages
auto_scroll = true                # Auto-scroll to newest messages
notifications = true              # Desktop notifications
show_turn_duration = true         # Agent turn execution duration
verbose = false                   # Verbose debug output
auto_compact = false              # Auto-compact conversation history

Key	Type	Default	Env Var	Description
theme	string	"dark"	OPSCODE_THEME	Active color theme
show_scrollbar	bool	false	—	Show vertical scrollbar in chat pane
show_timestamps	bool	true	—	Show timestamps on messages
auto_scroll	bool	true	—	Auto-scroll to newest streaming tokens
notifications	bool	true	—	Enable desktop notifications
show_turn_duration	bool	true	—	Show agent turn duration timer
verbose	bool	false	OPSCODE_VERBOSE	Enable verbose debugging mode
auto_compact	bool	false	—	Automatically compact conversation history

---

Tools

[tools]
shell_allow_list = ["terraform", "tofu", "kubectl", "helm", "ansible-playbook"]

Key	Type	Env Var	Description
shell_allow_list	list[string]	SHELL_ALLOW_LIST	Shell commands allowed without approval

Shell allow list values

The --shell-allow-list CLI flag (or shell_allow_list config) accepts:

• "recommended" — Curated set of safe, read-only commands (ls, cat, grep, kubectl get, terraform validate, etc.)
• "all" — Allow all shell commands without approval prompt
• Comma-separated list — Specific command binaries (e.g., "terraform,tofu,kubectl,helm")

---

Interpreter

[interpreter]
enable_interpreter = false
ptc = "safe"
ptc_acknowledge_unsafe = false

Key	Type	Default	Description
enable_interpreter	bool	false	Enable QuickJS JavaScript code interpreter (js_eval)
ptc	string	(none)	Programmatic Tool Calling (PTC) mode: "safe", "all", or comma-separated tools
ptc_acknowledge_unsafe	bool	false	Explicitly acknowledge unsafe PTC exposure

---

Permissions

Default permission settings for tool operations:

[permissions]
shell_read = true       # Allow non-mutating shell commands
shell_write = false     # Allow mutating shell commands
file_read = true        # Allow file read operations
file_write = false      # Allow file write / edit operations
infra_plan = false      # Allow infrastructure plan operations (terraform plan, etc.)
infra_apply = false     # Allow infrastructure apply operations (terraform apply, etc.)

Key	Type	Default	Description
shell_read	bool	true	Non-mutating shell commands (e.g., kubectl get, cat)
shell_write	bool	false	Mutating shell commands (e.g., kubectl apply, rm)
file_read	bool	true	File read operations (read_file, grep, glob, ls)
file_write	bool	false	File write/edit/delete operations
infra_plan	bool	false	Infrastructure dry-run operations (e.g., terraform plan)
infra_apply	bool	false	Infrastructure mutating operations (e.g., terraform apply)

tip

Use /permissions inside an interactive session to inspect and toggle these permissions dynamically.

---

Startup

[startup]
yolo_switcher = false     # Allow YOLO mode switching via Shift+Tab

---

Warnings

Suppress recurring informational banners:

[warnings]
suppress = ["yolo"]       # Suppress "YOLO is active" toast

---

Complete example

[model]
default = "anthropic:claude-opus-4-7"
reasoning_effort = "high"

[providers.openai]
enabled = true

[providers.anthropic]
enabled = true

[providers.ollama]
enabled = true
base_url = "http://localhost:11434"
models = ["llama3.3", "qwen2.5-coder"]

[ui]
theme = "dark"
show_timestamps = true
auto_compact = false

[interpreter]
enable_interpreter = false

[tools]
shell_allow_list = ["tofu", "terraform", "kubectl", "helm"]

[permissions]
shell_read = true
shell_write = false
file_read = true
file_write = false
infra_plan = false
infra_apply = false