Build¶
This section is for developers who build things that run on AIGr.id: AI models and services (blocks), multi-step workflows (vDAGs), and the policies that control them.
What do you want to build?¶
| I want to... | Read |
|---|---|
| Write the code that runs inside a block | Instance SDK |
| Control how my block load-balances, scales and handles failures | Block runtime |
| Plug in an existing system (vLLM, Triton, ...) instead of rewriting it | Third-party systems |
| Serve an LLM on one node (one or many GPUs) | LLM SDK utilities |
| Serve a model too big for one node | Model splitting |
| Publish a multi-step workflow with quotas and health checks | vDAG controller |
| Let applications call my blocks | Inference server |
| Customize placement, governance, scaling or anything else | Policies |
How the pieces fit¶
flowchart LR
SDK["Instance SDK<br/>(+ LLM utilities)"] -->|"container image"| COMP["Component"]
COMP -->|"deployed as"| BLK["Block<br/><small>executor, auto-scaler,<br/>health checker</small>"]
TP["Third-party system"] -. "optional backend" .-> BLK
BLK --> VDAG["vDAG + controller"]
BLK --> IS["Inference server"]
POL{{"Policies"}} -.-> BLK
POL -.-> VDAG
New to building on AIGr.id? Follow the Block creator guide first. It walks through the whole path from code to a running block.
Reference you'll use often¶
- Component spec and Block spec: how to describe and deploy your code
- vDAG spec: how to describe a workflow
- Metrics system: the metrics your policies can read