features settings

Settings

The settings feature manages ACP agent configuration, data source providers, and application preferences. It lives in frontend/src/features/settings/ (3 files).

Architecture Overview

graph TD
    A[SettingsPage] --> B[AgentStatusList]
    A --> C[Preferences Section]
    A --> D[DataSourcesSection]
    C --> E[Custom ACP Command Input]
    C --> F[Save Settings Button]
    D --> G[Provider Registry]
    G --> H[ProviderRow]
    H --> I[EnabledToggle]
    H --> J[API Key Input]
    H --> K[Test / Clear Buttons]

Loading

SettingsPage

frontend/src/features/settings/SettingsPage.tsx renders the settings page with three numbered sections:

Hero Section

A marketing-style header with "Settings & configuration." title and feature badges (ACP-compatible agents, Local execution, Source-backed research).

Section 01: Agents

Displays detected ACP agents via AgentStatusList. Includes a description explaining that unavailable agents should be checked against PATH or environment overrides (CODEX_ACP_BIN, INFI_CUSTOM_AGENT).

Section 02: Preferences

Renders the overrides form:

• Custom ACP command — a text input for an absolute path to a custom ACP agent binary. Stored in settings.custom_agent_command.
• Save button — persists settings via useUpdateSettings mutation, which calls update_settings on the backend.

The component uses a local copy of settings (localSettings) to avoid optimistic update issues. Changes are only persisted when the user clicks Save.

Section 03: Data Sources

Rendered by DataSourcesSection (see below).

AgentStatusList

frontend/src/features/settings/AgentStatusList.tsx displays a list of detected ACP agents in a card. Each row shows:

• Zero-padded index
• Agent logo (loaded via getLogoPath(agent.label))
• Agent label and command path (or "Not found on PATH")
• Status tag: green "Available" or gray "Missing"

If no agents are detected, shows a message: "No ACP agents detected on this machine."

DataSourcesSection

frontend/src/features/settings/DataSourcesSection.tsx (385 lines) manages the data source provider registry. It's the most complex settings component.

Provider Registry

Sources are grouped by category and displayed in a fixed order:

Category	Label
web_search	Web Search
filings	Filings
fundamentals	Fundamentals
market_data	Market Data
news	News
forums	Forums
screener	Screener

Each category shows a header with index, label, and enabled/total count (e.g., "03 / 05").

ProviderRow

Each provider row displays:

• Info column:
  • Key status: "Key stored", "Key required", or "No key"
  • Rate limit hint (if any)
  • Provider display name (16px, semibold)
  • Description
  • Links: Docs, Get key (if applicable), provider ID
• Toggle column: an EnabledToggle button (blue when enabled, gray when disabled)
• Key management (for providers requiring keys):
  • Password input for API key
  • Save/Replace button
  • Test button (validates the key)
  • Clear button (removes stored key)
  • Test result indicator (green "OK" or red "FAIL · reason")
• Reachability test (for keyless providers):
  • "Test reachability" button with result indicator

API Key Management Flow

sequenceDiagram
    participant U as User
    participant DS as DataSourcesSection
    participant CMD as commands.ts
    participant BE as Rust Backend
    participant KC as OS Keychain

    U->>DS: Paste API key
    U->>DS: Click "Save"
    DS->>CMD: setSourceKey(providerId, key)
    CMD->>BE: Tauri IPC: set_source_key
    BE->>KC: Store in OS keychain
    BE-->>CMD: OK
    DS->>CMD: refreshSourceKeyStatus()
    DS->>DS: Clear draft input

    U->>DS: Click "Test"
    DS->>CMD: testSourceKey(providerId)
    CMD->>BE: Tauri IPC: test_source_key
    BE-->>CMD: { status, message }
    DS->>DS: Show result for 4 seconds

    U->>DS: Click "Clear"
    DS->>CMD: clearSourceKey(providerId)
    CMD->>BE: Tauri IPC: clear_source_key
    BE->>KC: Remove from OS keychain

Loading

API keys are stored in the OS keychain — Infi never writes them to disk. The has_key field on SourceDescriptor indicates whether a key is currently stored.

Enable/Disable Toggle

Toggling a source calls setEnabledSources with the full list of enabled source IDs. This persists globally — the source will be available (or unavailable) for all future runs. Per-run overrides are handled by SourcesPopover in the research composer.

Test Results

Test results are displayed inline for 4 seconds, then auto-cleared. The result shows either a green "OK" or a red "FAIL · {first 40 chars of error message}".

Settings Model

The backend settings model is defined in src/infra/app_config.rs. The frontend AppSettings type includes:

Field	Type	Purpose
custom_agent_command	string | null	Absolute path to custom ACP agent binary
custom_agent_args	string[]	Arguments for the custom agent
timeout_secs	number	Agent run timeout
source_freshness_days	number	How old source data can be before flagged stale
disclaimer	string	Report disclaimer text
model_by_agent	Record<string, string>	Per-agent model override map
enabled_sources	string[]	List of enabled source provider IDs
sources_with_keys	string[]	List of source IDs that have stored keys

Settings are fetched via useSettings() (TanStack Query) and updated via useUpdateSettings() mutation.

Agent Selection and Model Override

Agent and model selection happens in two places:

1. Settings page — the AgentStatusList shows detected agents but doesn't allow selection.
2. Research composer — the AgentSelector dropdown lets users pick an agent and optionally override its model. The selection is persisted to the global store (agentId, modelByAgent) and to the backend via persistModelByAgent().

The modelByAgent map is loaded from settings on app start and updated whenever the user changes a model selection.

Key Source Files

File	Lines	Purpose
frontend/src/features/settings/SettingsPage.tsx	185	Main settings page with 3 sections
frontend/src/features/settings/DataSourcesSection.tsx	385	Data source provider registry with key management
frontend/src/features/settings/AgentStatusList.tsx	62	ACP agent status display

Related Pages

• Frontend Architecture — state management and TanStack Query patterns
• Run Analysis — where agent selection and source configuration are used
• Portfolio — portfolio analysis uses the same agent and source configuration