Config Reference

This reference lists the Open Interpreter settings users most often need. The source of truth for every field is the generated schema in codex-rs/core/config.schema.json.

Top-Level Keys

Key	Values	Purpose
`model`	string	Default model id.
`model_provider`	provider id	Provider entry used for `model`.
`model_reasoning_effort`	`minimal`, `low`, `medium`, `high`, `xhigh`	Reasoning budget for supported models.
`model_reasoning_summary`	`auto`, `concise`, `detailed`, `none`	How much reasoning summary to show.
`model_verbosity`	`low`, `medium`, `high`	Verbosity for supported Responses API models.
`approval_policy`	`untrusted`, `on-request`, `never`	When to ask before running commands.
`approvals_reviewer`	`user`, `auto_review`	Who reviews eligible approval prompts.
`sandbox_mode`	`read-only`, `workspace-write`, `danger-full-access`	Local command sandbox.
`default_permissions`	profile name	Selects a beta permissions profile.
`personality`	`friendly`, `pragmatic`, `none`	TUI communication style.
`web_search`	`cached`, `live`, `disabled`	Web search behavior.
`log_dir`	path	Directory for logs.
`file_opener`	`vscode`, `vscode-insiders`, `windsurf`, `cursor`, `none`	Editor used for file citations.
`harness`	string	Open Interpreter harness compatibility mode.
`harness_guidance`	boolean	Allow OI guidance inside harness mode.
`check_for_update_on_startup`	boolean	Enable managed standalone update checks.

Feature Flags

Key	Default posture	Purpose
`features.apps`	off	App/connectors surface.
`features.plugins`	off	Plugin bundles.
`features.hooks`	on	Lifecycle hooks.
`features.memories`	off	Persistent memory generation/use.
`features.multi_agent`	on	Subagent tools and `/agent`.
`features.shell_tool`	on	Built-in shell command tool.
`features.shell_snapshot`	on	Shell environment snapshotting.
`features.unified_exec`	on except where unsupported	PTY-backed exec tool.
`features.undo`	off	Undo support where available.
`features.network_proxy`	off	Sandboxed network proxy controls.

Provider Tables

[model_providers.example]
name = "Example"
base_url = "https://api.example.com/v1"
env_key = "EXAMPLE_API_KEY"
wire_api = "responses"
request_max_retries = 4
stream_max_retries = 5
stream_idle_timeout_ms = 300000

Provider auth can also be command-backed:

[model_providers.example.auth]
command = "example-token"
args = ["print"]
refresh_interval_ms = 300000
timeout_ms = 5000

Sandbox Tables

[sandbox_workspace_write]
network_access = false
exclude_tmpdir_env_var = false
exclude_slash_tmp = false
writable_roots = ["/tmp/project-cache"]

For finer-grained access, use permissions profiles instead of mixing both systems in one active config.

Permission Profiles

default_permissions = "project-edit"

[permissions.project-edit.filesystem]
":minimal" = "read"

[permissions.project-edit.filesystem.":workspace_roots"]
"." = "write"
"**/*.env" = "deny"

[permissions.project-edit.network]
enabled = true

[permissions.project-edit.network.domains]
"api.openai.com" = "allow"
"*.github.com" = "allow"

See Permissions for the complete model.

MCP Tables

[mcp_servers.docs]
command = "docs-mcp"
args = ["--stdio"]
env = { DOCS_TOKEN = "env:DOCS_TOKEN" }
startup_timeout_sec = 10
tool_timeout_sec = 60
required = false
enabled_tools = ["search", "read"]
disabled_tools = ["delete"]
default_tools_approval_mode = "prompt"

[mcp_servers.docs.tools.search]
approval_mode = "approve"

HTTP MCP servers use url, bearer_token_env_var, http_headers, and env_http_headers.

Hooks

Hooks can be configured inline:

[[hooks.PreToolUse]]
matcher = "^Bash$"

[[hooks.PreToolUse.hooks]]
type = "command"
command = "python3 .openinterpreter/hooks/pre_tool_use.py"
timeout = 30
statusMessage = "Checking command"

Or in hooks.json next to the active config layer. See Hooks.

Agents

[agents]
max_threads = 6
max_depth = 1
job_max_runtime_seconds = 1800

[agents.explorer]
description = "Investigate code and report findings without editing."
model_reasoning_effort = "medium"
sandbox_mode = "read-only"

See Subagents.

Give feedback