Knowledge Base — the Qdrant vector store

A Qdrant vector store (MCP server qdrant-vector-db) is the agents' shared memory.

They semantic-search it for grounding — real docs, real code, real specs — before writing, so answers are based on FT reality, not guesses.

Read-only retrieval (RAG); nothing is mutated by a query.

kb-map

What’s in it — who uses it — when

Collection What’s in it (≈ size) Who When

ft_docs_v2 ⭐

Product Antora docs — 17 sources (ft-acs, ft-device-network-service, oneiot-ui, ft-qoe-web, northbound/service/provision, emulators, grafana…). ~8k

ft-developer, coding agents, doc-writer, bugfix

Understand a feature/flow and ground a change. Scope with source_filter=<product>. Primary docs source.

ft_code

Indexed source code (re-indexed on merge to main). ~18k

coding agents, ft-developer, audit

Find an existing implementation / nearest analogue before writing new code.

bbf_params

BBF/OMA data-model parameters — TR-181 / TR-098 / USP paths & types. ~22k

coding agents, ft-developer, bugfix, cwmp-xml / oma-lwm2m-expert

Validate a protocol parameter path before coding/testing (protocol = source of truth).

bbf_protocol_specs

BBF/OMA protocol spec text — TR-069 / TR-369 / LwM2M messages & semantics. ~530

same as bbf_params

Confirm message shape / semantics against the spec; derive test fixtures.

ft_bugs

Historical bugs & issues. ~58k

bugfix-agent, coding agents

Check whether an issue occurred before and how it was fixed.

ft_ui_map

UI screen / component / flow map. ~280

ft-angular-coding-agent, doc-writer

Locate a UI screen or flow when changing/documenting the frontend.

ft_qa_notes

QA notes (nascent). ~7

bugfix-agent, QA flows

QA context — small for now; will grow.

ft_docs

Older docs, superseded by ft_docs_v2. ~5k

fallback only

Only when ft_docs_v2 lacks the topic.

How agents query it (rules)

Rule Detail

Scope to a product

pass source_filter=<product> on semantic_search so results stay on the repo in hand (sources via list_sources).

Ground before writing

coding agents search ft_docs_v2 + ft_code for the nearest analogue before new code.

Protocol = truth

any TR-069/181/369 / LwM2M parameter is validated against bbf_params / bbf_protocol_specs first.

Fresh code

ft_code is re-indexed on merge to main (deploy/ci/ft-code-reindex.yml) so search reflects HEAD.

Read-only

retrieval only — the KB is never written by a task; it’s grounding, not state.
For current library/framework docs (Spring, Angular, NgRx, etc.) agents use context7, not this KB — Qdrant holds FT-specific knowledge (our docs, code, bugs, protocol data), context7 holds external libs.

This is the read-only knowledge layer. The agents' writeable memory — the root causes and decisions they author at the end of a task — lives in the Obsidian Knowledge Vault. Qdrant is the library you look things up in; the vault is the lab notebook the team writes.
Agents reach it via the qdrant-vector-db MCP tools (list_collections, list_sources, semantic_search, …). See the agent fleet for who runs what.