Configuration

intake works without configuration — it only needs an LLM API key. To customize behavior, use an .intake.yaml file at the root of your project.

---

Loading priority

Configuration is loaded in layers. Each layer overrides the previous one:

CLI flags  >  .intake.yaml  >  preset  >  defaults

1. Defaults: default values in the code
2. Preset: if --preset is used, a predefined set is applied
3. .intake.yaml: project configuration file
4. CLI flags: command-line options always win

---

.intake.yaml file

Create an .intake.yaml file at the root of your project. Full example:

# LLM model configuration
llm:
  model: claude-sonnet-4         # Any model supported by LiteLLM
  api_key_env: ANTHROPIC_API_KEY  # Environment variable with the API key
  max_cost_per_spec: 0.50         # Maximum budget per spec (USD)
  temperature: 0.2                # 0.0 = deterministic, 1.0 = creative
  max_retries: 3                  # Retries on failure
  timeout: 120                    # Timeout per LLM call (seconds)

# Project configuration
project:
  name: my-project                # Name (auto-detected if empty)
  stack: []                       # Technology stack (auto-detected if empty)
  language: en                    # Language of generated content
  conventions: {}                 # Custom conventions (key: value)

# Spec configuration
spec:
  output_dir: ./specs             # Where to save generated specs
  requirements_format: ears       # ears | user-stories | bdd | free
  design_depth: moderate          # minimal | moderate | detailed
  task_granularity: medium        # coarse | medium | fine
  include_sources: true           # Include source traceability
  version_specs: true             # Create versioned directories
  generate_lock: true             # Generate spec.lock.yaml
  risk_assessment: true           # Include risk assessment
  auto_mode: true                 # Auto-detect quick/standard/enterprise

# Verification configuration
verification:
  auto_generate_tests: true       # Generate acceptance checks
  test_output_dir: ./tests/generated
  checks: []                      # Additional custom checks
  timeout_per_check: 120          # Timeout per check (seconds)

# Export configuration
export:
  default_format: generic         # architect | claude-code | cursor | kiro | generic
  architect_include_guardrails: true
  architect_pipeline_template: standard
  claude_code_generate_claude_md: true

# Connectors (preparation for Phase 2)
connectors:
  jira:
    url: ""                       # Jira base URL
    email: ""                     # Authentication email
    api_token_env: JIRA_API_TOKEN # Environment variable with the API token
  confluence:
    url: ""                       # Confluence base URL
    email: ""                     # Authentication email
    api_token_env: CONFLUENCE_API_TOKEN  # Environment variable with the API token
  github:
    token_env: GITHUB_TOKEN       # Environment variable with the token

# Security
security:
  redact_patterns: []             # Regex patterns to redact from output
  redact_files:                   # Files to never include
    - "*.env"
    - "*.pem"
    - "*.key"

---

Complete field reference

llm section

Field	Type	Default	Description
model	string	claude-sonnet-4	LLM model. Any model that LiteLLM supports.
api_key_env	string	ANTHROPIC_API_KEY	Name of the environment variable containing the API key.
max_cost_per_spec	float	0.50	Maximum budget per spec in USD. If exceeded, analysis stops.
temperature	float	0.2	Model temperature. Lower = more deterministic.
max_retries	int	3	Number of retries on LLM failures.
timeout	int	120	Timeout per LLM call in seconds.

Supported models:

Provider	Examples	Environment variable
Anthropic	claude-sonnet-4, claude-opus-4, claude-haiku-4-5	ANTHROPIC_API_KEY
OpenAI	gpt-4o, gpt-4, gpt-3.5-turbo	OPENAI_API_KEY
Google	gemini/gemini-pro, gemini/gemini-flash	GEMINI_API_KEY
Local	ollama/llama3, ollama/mistral	(no key needed)

project section

Field	Type	Default	Description
name	string	""	Project name. If empty, it is generated from the init description.
stack	list[string]	[]	Technology stack. If empty, it is auto-detected from project files.
language	string	en	Language for generated content (e.g.: en, es, fr).
conventions	dict	{}	Project conventions as key-value pairs.

Stack auto-detection looks for 28+ marker files in the project directory:

File	Detected stack
package.json	javascript, node
tsconfig.json	typescript
pyproject.toml	python
Cargo.toml	rust
go.mod	go
pom.xml	java, maven
Dockerfile	docker
next.config.js	nextjs
…	…

It also inspects the contents of pyproject.toml and package.json to detect frameworks (fastapi, django, react, vue, etc.).

spec section

Field	Type	Default	Description
output_dir	string	./specs	Output directory for specs.
requirements_format	string	ears	Requirements format. Options: ears, user-stories, bdd, free.
design_depth	string	moderate	Level of design detail. Options: minimal, moderate, detailed.
task_granularity	string	medium	Task granularity. Options: coarse, medium, fine.
include_sources	bool	true	Include sources.md with requirement-to-source traceability.
version_specs	bool	true	Create versioned subdirectories for specs.
generate_lock	bool	true	Generate spec.lock.yaml with hashes and metadata.
risk_assessment	bool	true	Run risk assessment (additional LLM phase).
auto_mode	bool	true	Auto-detect generation mode (quick/standard/enterprise) based on source complexity. Ignored if --mode is used in the CLI.

Generation modes:

Mode	Auto-detection criteria	Generated files
quick	<500 words, 1 source, no structure	context.md + tasks.md
standard	Everything that is not quick or enterprise	All 6 complete spec files
enterprise	4+ sources OR >5000 words	All 6 files + detailed risks

Requirements formats:

Format	Description	Best for
ears	Easy Approach to Requirements Syntax. Structured format with conditions.	Formal specifications
user-stories	”As a [role], I want [action] so that [benefit]”.	Agile teams
bdd	Given/When/Then. Behavior-driven development.	Acceptance tests
free	Free format. No imposed structure.	Quick prototypes

Design levels:

Level	Description
minimal	Only main components and critical decisions.
moderate	Components, files, technical decisions, and dependencies.
detailed	All of the above plus interaction diagrams, edge cases, and performance considerations.

Task granularity:

Level	Description
coarse	Large, few tasks. Each task covers a complete component.
medium	Balance between granularity and quantity.
fine	Small, atomic tasks. Each task is ~15-30 minutes of work.

verification section

Field	Type	Default	Description
auto_generate_tests	bool	true	Automatically generate acceptance checks from requirements.
test_output_dir	string	./tests/generated	Directory for generated tests.
checks	list[string]	[]	Additional custom checks.
timeout_per_check	int	120	Maximum timeout per individual check in seconds.

export section

Field	Type	Default	Description
default_format	string	generic	Default export format. Options: architect, claude-code, cursor, kiro, generic.
architect_include_guardrails	bool	true	Include guardrails in architect pipelines.
architect_pipeline_template	string	standard	Pipeline template for architect.
claude_code_generate_claude_md	bool	true	Generate CLAUDE.md when exporting for Claude Code.

connectors section (preparation)

Configuration for direct API connectors. Connectors are not yet implemented (they will arrive in a future version), but the configuration infrastructure is already in place.

Field	Type	Default	Description
jira.url	string	""	Base URL of the Jira instance.
jira.email	string	""	Email for Jira authentication.
jira.api_token_env	string	"JIRA_API_TOKEN"	Environment variable with the Jira API token.
confluence.url	string	""	Base URL of the Confluence instance.
confluence.email	string	""	Email for Confluence authentication.
confluence.api_token_env	string	"CONFLUENCE_API_TOKEN"	Environment variable with the Confluence API token.
github.token_env	string	"GITHUB_TOKEN"	Environment variable with the GitHub token.

Note: Currently, when using URIs like jira://PROJ-123 in -s, intake shows a warning indicating that the connector is not available. In the meantime, export the data from the web interface and use JSON files.

security section

Field	Type	Default	Description
redact_patterns	list[string]	[]	Regex patterns that will be removed from generated content.
redact_files	list[string]	["*.env", "*.pem", "*.key"]	Glob patterns of files that will never be included.

---

Presets

Presets are predefined configurations for common use cases. They are applied with --preset:

intake init "My feature" -s reqs.md --preset minimal

Comparison

Field	minimal	standard	enterprise
Use case	Quick prototype	Normal teams	Regulated / critical
max_cost_per_spec	$0.10	$0.50	$2.00
temperature	0.3	0.2	0.1
requirements_format	free	ears	ears
design_depth	minimal	moderate	detailed
task_granularity	coarse	medium	fine
include_sources	false	true	true
risk_assessment	false	true	true
generate_lock	false	true	true

When to use each preset

• minimal: Quick prototyping, exploratory ideas, solo developer. Low cost, minimal output.
• standard: The default option. Good balance between detail and cost for teams of 2-5 people.
• enterprise: For large teams, regulated projects, or when complete traceability and exhaustive risk assessment are needed.

---

Environment variables

intake looks for these environment variables for LLM provider authentication:

Variable	Provider	Example
ANTHROPIC_API_KEY	Anthropic (Claude)	sk-ant-api03-...
OPENAI_API_KEY	OpenAI (GPT)	sk-...

Set the variable according to your provider:

# Anthropic
export ANTHROPIC_API_KEY=sk-ant-api03-your-key-here

# OpenAI
export OPENAI_API_KEY=sk-your-key-here

If you use a different provider, configure llm.api_key_env in .intake.yaml:

llm:
  model: gemini/gemini-pro
  api_key_env: GEMINI_API_KEY

---

Generate config automatically

If you don’t have an .intake.yaml, intake uses sensible defaults. To create a basic configuration file:

intake doctor --fix

This creates a minimal .intake.yaml that you can customize:

# intake configuration
llm:
  model: claude-sonnet-4
  # max_cost_per_spec: 0.50
project:
  name: ""
  language: en
  # stack: []
spec:
  output_dir: ./specs