General & Pipeline

Configuring [general], [pipeline], and [knowledge] tables in niki.toml.

[general], [pipeline], and [knowledge]

Core operational settings for pipeline limits and context ingestion.

[general] Table

[general]
max_revision_rounds = 3      # Maximum Reviewer -> Coder feedback iterations
output_dir          = ".niki" # Task output root; tasks land under tasks/<id>/
spend_cap_usd       = 5.0    # Hard estimated-dollar ceiling per run
max_diff_lines      = 200    # Reviewer nudge + report warning; not a hard rejection
max_context_chars   = 48000  # Planner context budget

[pipeline] Table

Customize the execution topology and stages:

[pipeline]
topology                    = "auto" # "auto" | "multiagent" | "singleagent"
single_agent_max_complexity = "Low"  # At or below this level, auto uses Planner + solo Coder
max_revision_rounds         = 3

stages = [
  { role = "planner",  provider = "anthropic", model = "claude-sonnet-4-20250514" },
  { role = "coder",    provider = "anthropic", model = "claude-sonnet-4-20250514" },
  { role = "tester",   provider = "openai",    model = "gpt-4o-mini" },
  { role = "reviewer", provider = "anthropic", model = "claude-sonnet-4-20250514" }
]

auto is the default topology. Security/parallel configuration and High/Security risk can force multiagent; use topology = "multiagent" when an explicit stage list must always retain the full chain.

[knowledge] Table

Include extra project documentation or remote documentation pages into agent prompts:

[knowledge]
doc_globs        = ["docs/**/*.md", "README.md", "ARCHITECTURE.md"]
urls             = ["https://example.com/api-spec.md"]
max_source_chars = 8000 # Truncation limit per document