falk uses files, not databases. All configuration is version-controlled and PR-reviewable.
| File | Purpose | Loaded when | Who edits |
|---|---|---|---|
falk_project.yaml |
LLM settings, session config, quick context | Startup | Platform team |
semantic_models.yaml |
Metrics, dimensions, data structure | Startup | Data engineers |
RULES.md |
Agent behavior, tone, formatting | Startup (included in system prompt) | Anyone |
knowledge/business.md |
Business terms, glossary, company context | Startup (if enabled) | Domain experts |
knowledge/gotchas.md |
Data quality issues, caveats | Startup (if enabled) | Data stewards |
.env |
API keys, secrets | Startup | DevOps |
| Question | File |
|---|---|
| “What LLM should we use?” | falk_project.yaml |
| “Revenue = SUM(orders.amount)” | semantic_models.yaml |
| “Always use nested bullets” | RULES.md |
| “MRR = monthly recurring revenue” | knowledge/business.md |
| “Revenue delayed 24h” | knowledge/gotchas.md |
| “Be conversational, not robotic” | RULES.md |
| Example questions | falk_project.yaml |
| Custom project tools | falk_project.yaml + Python modules |
See Context for detailed guidance.
semantic_models.yaml → "What data exists" (metrics, dimensions)
falk_project.yaml → "How the agent runs" (LLM, session store)
RULES.md → "How the agent behaves" (always included)
knowledge/ → "Deep domain knowledge" (startup when enabled)
Update any config file and the agent picks up the changes automatically — no prompt editing needed.