AI Platform Overview
The Oridecon AI platform is 17 packages. oridecon-ai is the orchestrator: it discovers the others through entry points and wires them in the container. Application code talks to protocols in oridecon-contracts, not to a concrete SDK.
All of it is alpha (0.1.x), MIT, in dbtinoy-/oridecon under experimental/ai/ (branch dev). Pin versions. Add AI after a working HTTP app — see Adoption Paths.
uv add oridecon-clioridecon new project my-app --template web-apiuv add oridecon-ai-llmConfig lives in application.yaml under ai: / per-package keys. Layers and the full catalog: AI Architecture.
1. Principles
Section titled “1. Principles”- Protocol-first. LLM clients, RAG, agents, and memory implement contracts in
oridecon.contracts.ai. - Never import each other.
oridecon-ai-agentsdoes not importoridecon-ai-rag. The container injects whatever is registered. - Orchestrator discovers.
oridecon-aifinds sub-packages via entry points. You install only the ones you use.
2. The 17 packages
Section titled “2. The 17 packages”| Layer | Packages |
|---|---|
| Orchestrator | oridecon-ai |
| Base | oridecon-ai-llm |
| Knowledge | oridecon-ai-rag, oridecon-vector |
| Reasoning | oridecon-ai-agents, oridecon-ai-skills |
| Memory | oridecon-ai-memory, oridecon-ai-session |
| Integration | oridecon-ai-mcp |
| Infrastructure | oridecon-ai-workers, oridecon-ai-observability, oridecon-ai-feedback |
| Policy & quality | oridecon-ai-guard, oridecon-ai-governance, oridecon-ai-evaluation, oridecon-ai-prompt |
| Relay | oridecon-ai-relay, oridecon-ai-relay-gateway |
oridecon-vector is a general-purpose package (also used outside AI). It still talks through contracts.
3. Shared value types
Section titled “3. Shared value types”Immutable types live in oridecon.contracts.ai — do not invent a second copy in an extension.
from oridecon.contracts.ai.llm import ChatMessage, Role
msg = ChatMessage( role=Role.USER, content="How does Result[T, E] work?",)| Role | Meaning |
|---|---|
SYSTEM | Instructions for the model |
USER | Input from the human |
ASSISTANT | Response from the model |
TOOL | Output from a function call |
Inject the protocol on the constructor. Domain failures are Result. Copy examples/support-agent rather than resolving the container inside a service.
Next Steps
Section titled “Next Steps”- AI Architecture — layers and the never-import rule
- AI agents — tools and strategies
- RAG — chunk, retrieve, cite
- Examples —
support-agentand the rest of the fleet - Compatibility — extras (
[openai],[anthropic], …)