Architecture
The two paths through Doram, its components, and where your data lives.
Company meaning is discovered, validated and approved before anyone asks a question. When a question comes in, Doram only looks it up.
Two paths
The write path builds the second brain: Discover → Validate → Approve. See Company second brain.
The read path answers questions: Understand → Resolve → Compile → Execute. See How Doram answers.
The two paths meet only at the graph. Nothing on the read path can write to the graph, and nothing on the write path can answer a question.
Components
| Component | What it does | Uses an LLM |
|---|---|---|
| Connectors | Read your sources with read-only access | No |
| Crawler agent | Drafts facts from evidence | Yes |
| Validation | Tests each draft | No |
| Approval | Approves by risk, or sends to the owner | No |
| Monitoring agent | Turns source changes into new drafts | No |
| Feedback agent | Turns usage into new drafts | No |
| Retriever | Turns question parts into a context pack | No |
| Semantic compiler | Turns a context pack into SQL | No |
| AI analyst | Reads the question and explains the answer | Yes |
The LLM proposes, and code decides. The LLM never writes SQL and never sets a trust level.
Where things live
| What | Where |
|---|---|
| Your data | Stays in your warehouse. Queries run there through a read-only account. |
| The second brain | A versioned graph in Postgres: definitions, relationships, profiles and evidence. No rows of your data. |
The retriever, compiler and analyst hold no state and scale horizontally. Each fact is written in one transaction, and after the first run, connectors read only what changed.