tulipfarm docsConcepts
How TulipFarm's parts fit together — soul, resources, agents, skills, knowledge, LLM tiers, and secrets.
You operate TulipFarm by talking to it. You describe what you want, and agents build and run it for you — see Building by chat. Underneath, two things make that work, and the separation between them explains everything else:
- Definitions live in the soul (a git repository): resource schemas, agents, skills. These are files agents write when you ask them to — readable, diffable, and versioned.
- Data lives in the datastore (PostgreSQL): the records your resources hold, your knowledge base, chats, and secrets.
Building the soul changes what your instance is. Writing to the datastore changes what it knows. Agents do both — but the two are governed differently. Soul writes are never gated; real-world actions an agent takes are gated by its autonomy level.
The parts
Building by chat
How you build your system — describe it, and agents build it for you.
Soul
The git-backed configuration store that defines your instance.
Resources
Schema-defined data types and the records that fill them.
Agents
The actors that run turns, defined by an AGENT.md file.
Skills
Reusable instructions agents load on demand, audited before install.
Knowledge
The searchable corpus agents draw on, backed by vector search.
LLM tiers
How TulipFarm chooses and falls back across model providers.
Secrets
Encrypted storage for the credentials your instance needs.