Agents
The five Datapace agents (FinOps, Performance, Quality, Migration, Documentation), what each proposes on the graph, what a proposal contains, who approves.
Five agents, one graph
Datapace's agents work on the confirmed graph. Each one owns a workstream, reads the evidence the graph holds for it, and drafts proposals. None of them executes anything.
| Agent | What it proposes | The evidence it reads |
|---|---|---|
| FinOps | Right-sizing, idle databases, storage growth, upgrades priced as risk | Instance class, cost, usage, freshness |
| Performance | Index and rewrite work for slow queries, overdue engine upgrades, regressions tied to the change that caused them | Workload signals, plan shapes, engine versions |
| Quality | Checks on the confirmed definitions, drift between the schema and the graph | Semantics, constraints, sample structures |
| Migration | Field-level mappings from a source system into a target model, with findings before load | The source graph, the target model, lineage |
| Documentation | Updates to the confirmed graph as the estate changes | New objects, changed workloads, re-flagged columns |
What a proposal contains
A proposal is a change with its case attached. Whoever reviews it should not need to open another tool. An illustrative proposal from the FinOps agent:
proposal: PR-0418
agent: FinOps
change: resize analytics-replica from db.r6g.2xlarge to db.r6g.xlarge
affects:
- database: analytics-replica # Postgres, reporting
evidence:
- cpu p95 over 30 days: 11%
- connections p95: 14 of 400
- largest query, monthly close, runs on the primary
- last schema change: 2024-11
risk: low # a replica; reversible in one step
status: awaiting approval
approver: platform teamThe values are illustrative. The shape is the rule: what changes, what it affects, the evidence, the risk as the agent reads it, and where the proposal is in the approval flow.
Agents propose, people approve
A proposal waits for a person. Your team reviews it, approves it, rejects it, or asks for more. Approval is recorded with who gave it, and the outcome is recorded when the change lands. See policy, approval, and audit and reviewing a proposal.
The same rule holds for the Documentation agent: an update to the confirmed graph is a proposal like any other, and it enters the graph only once an expert accepts it.
Your model, our context
Datapace is the graph and the gate, never the model. The agents run on the language model your team chooses; the graph is what makes their proposals specific to your estate rather than generic advice. Bringing a different model changes nothing about the graph, the evidence, or the approval flow.