term-llm
Docs → Reference → Configuration

Config

Configuration

Inspect, edit, and understand the main config file, provider settings, search configuration, and per-command overrides.

On this page

Configuration commands

term-llm config
term-llm config edit
term-llm config path
term-llm config get default_provider
term-llm config set default_provider zen
term-llm config reset

The main config file lives at:

~/.config/term-llm/config.yaml

Configuration shape

A typical config has a few major parts:

  • default_provider for the global LLM default
  • providers for model-specific credentials and routing
  • per-command blocks such as exec, ask, and edit
  • feature-specific blocks such as image, embed, search, sessions, and tools

Example

default_provider: anthropic

providers:
  anthropic:
    model: claude-sonnet-4-6

  openai:
    model: gpt-5.2
    credentials: codex

  xai:
    model: grok-4-1-fast

  openrouter:
    model: x-ai/grok-code-fast-1
    app_url: https://github.com/samsaffron/term-llm
    app_title: term-llm

exec:
  suggestions: 3
  instructions: |
    I use Arch Linux with zsh.
    I prefer ripgrep over grep and fd over find.

ask:
  instructions: |
    Be concise. I'm an experienced developer.

edit:
  model: gpt-5.2-codex
  diff_format: auto

search:
  provider: duckduckgo

tools:
  max_tool_output_chars: 20000

Per-command overrides

Each command can override provider and model independently of the global default.

default_provider: anthropic

providers:
  anthropic:
    model: claude-sonnet-4-6
  openai:
    model: gpt-5.2
  zen:
    model: glm-4.7-free

exec:
  provider: zen
  model: glm-4.7-free

ask:
  model: claude-opus-4

edit:
  provider: openai
  model: gpt-5.2-codex

Precedence is:

  1. CLI flag such as --provider openai:gpt-5.2
  2. per-command config such as exec.provider or ask.model
  3. global provider selection via default_provider and providers.<name>.model

Sessions config

sessions:
  enabled: true
  max_age_days: 0
  max_count: 0
  path: ""

Use this to control whether sessions are persisted, how long they are kept, and where the SQLite database lives.

Search config

search:
  provider: perplexity
  force_external: false

  perplexity:
    api_key: ${PERPLEXITY_API_KEY}

  exa:
    api_key: ${EXA_API_KEY}

  brave:
    api_key: ${BRAVE_API_KEY}

Search is large enough to deserve its own page; see Search.

Image and embedding config

image:
  provider: gemini
  output_dir: ~/Pictures/term-llm

embed:
  provider: gemini

Each feature block can hold provider-specific credentials and defaults. The image and embedding providers are independent of the main text provider.

Dynamic secrets and endpoints

term-llm supports dynamic resolution for some config values:

  • op://... for 1Password secret references
  • srv://... for DNS SRV-based endpoint discovery
  • $() for command-based resolution

Example:

providers:
  production-llm:
    type: openai_compatible
    model: Qwen/Qwen3-30B-A3B
    url: "srv://_vllm._tcp.ml.company.com/v1/chat/completions"
    api_key: "op://Infrastructure/vLLM Cluster/credential?account=company.1password.com"

These values are resolved lazily when term-llm actually needs them.

Diagnostics

diagnostics:
  enabled: true

When edit retries fail, diagnostics can capture prompts, partial responses, and failure context for inspection.