Applied AI fellowship ↔ Astrocyte vocabulary
This note is for educators, certification authors, and architects who teach against vendor-neutral agentic-stack curricula (for example the Applied AI Fellowship) while building or reviewing Astrocyte deployments. It maps course vocabulary to framework primitives without requiring either side to rename concepts.
Plane alignment
Section titled “Plane alignment”| Fellowship / stack plane | Role | Astrocyte |
|---|---|---|
| Context & Memory | Governed memory and retrieval—not “RAG only” | Core retain / recall / reflect; hybrid retrieval; Tier 1 vs Tier 2 providers (Architecture framework §2) |
| Control (outside the agent loop) | Deterministic policy, identity surfaces, budgets | Per-bank AuthZ, policy layer, opaque principal from the caller; MCP tool graphs are owned by the harness, not Astrocyte (repository README non-goals) |
| Runtime | Orchestration, workers, sandboxes | Out of scope — agent frameworks own the harness (Architecture framework §1) |
| Integrations | MCP, search, APIs | MCP as an integration surface (MCP server); memory is invoked from the application loop |
| Observability | Traces, evals, cost | Benchmarks, in-process bank health (Bank health & utilization), Evaluation, Event hooks |
| Data | Tenancy, schemas, contracts | Memory banks and Tier 1/2 SPIs; optional Memory export sink for warehouse/lakehouse/history (Memory export sink, Storage & data planes); not a full enterprise data platform |
Memory framing
Section titled “Memory framing”| Fellowship concept | Meaning in teaching | In Astrocyte (approximate) |
|---|---|---|
| Memory ≠ RAG | Cognition-oriented persistence vs one-off retrieval | Built-in Tier 1 pipeline, policies, and reflect; not limited to chunk similarity |
| World / Experience / Opinion / Observation | Pedagogical “memory networks” | Model with multiple banks, use-case profiles, or provider semantics—there is no mandatory one-to-one schema in core |
| TEMPR / Hindsight (and similar named patterns) | Named retrieval / priming patterns in course materials | Achieved through structured retrieval, hybrid stores, and engine providers—not a single builtin product name in this repo |
| Harness vs context | Agent loop vs what the model sees | Same split as harness engineering vs context engineering in Architecture framework §1 |
Suggested teaching hooks
Section titled “Suggested teaching hooks”- Context & Memory week (architect track): Read Architecture framework §§1–3 and Design principles; optional hands-on: Astrocyte + a Tier 1 adapter (for example PostgreSQL + pgvector) with a written memory boundary spec (banks, principals, recall strategies).
- Capstone: Keep the fellowship artifact vendor-neutral; cite Astrocyte only as one example implementation of the governed memory boundary.
External links
Section titled “External links”- Applied AI Fellowship — syllabus, stack diagram, glossary.
- Astrocyte repository — source and releases.