AI-powered virtual executive team — a single coherent executive persona backed by 8 specialist agents (FastAPI + Next.js).
fix(knowledge): share one lean embedding session across chroma collections (#292)
Rufusb05a75b5f43f
diff --git a/packages/core/openexecutive/architecture/architecture-facts.yaml b/packages/core/openexecutive/architecture/architecture-facts.yamlindex 80f44739..b1109d40 100644--- a/packages/core/openexecutive/architecture/architecture-facts.yaml+++ b/packages/core/openexecutive/architecture/architecture-facts.yaml@@ -1190,6 +1190,25 @@ routing:carrying the matched term.knowledge:+ embedding_function: |+ Every ChromaDBStore collection is opened with one shared embedding+ function (knowledge.store._embedding_function): all-MiniLM-L6-v2 ONNX,+ the same model and vectors as Chroma's DefaultEmbeddingFunction.++ Why not Chroma's default: it builds a new ONNXMiniLM_L6_V2 per call, so+ every query reloads the model (~240 ms), and its session keeps ONNX+ Runtime's CPU memory arena, which retains the peak of a 32-text batch+ padded to 256 tokens. A 300-chunk upload peaked near 760 MB. The shared+ session is built once, runs with enable_cpu_mem_arena=False and embeds 4+ texts per run: ~300 MB peak, ~30 ms per query, ~90 MB more at idle+ because the model stays loaded.++ Compatibility invariants: collections persist the function they were+ created with ({"name": "default", "config": {}} on every existing+ install), and Chroma raises on a differently named function, so the+ shared one answers name() == "default" with an empty config. It must NOT+ be a DefaultEmbeddingFunction instance: Collection._embed ignores those+ and embeds with a fresh default rebuilt from the collection config.distance_gates: |Two gates. KNOWLEDGE_DISTANCE_THRESHOLD (default 0.55) is the shared one,applied to COMPANY, NOTION_WIKI and RECENT_RESEARCH.diff --git a/packages/core/openexecutive/architecture/prebuilt/rag.json b/packages/core/openexecutive/architecture/prebuilt/rag.jsonindex 44f55dc2..cf66d1f0 100644--- a/packages/core/openexecutive/architecture/prebuilt/rag.json+++ b/packages/core/openexecutive/architecture/prebuilt/rag.json@@ -1,7 +1,7 @@{"section_id": "rag","title": "Knowledge & RAG",- "markdown": "Retrieval is handled by `openexecutive.knowledge.retriever.retrieve`, querying ChromaDB collections through `openexecutive.knowledge.store.ChromaDBStore`. When the caller passes `record_source`, `retrieve` also names every document its block quotes — company uploads by filename, built-in knowledge by a title made from its filename, Notion pages, synced Drive files by name with their Drive link, research notes by date, and earlier published documents with a link to `/artifacts/<id>` — so the web chat can list what an answer looked at: the chat route's own search, and each specialist's that answered (see the Lifecycle and API sections). The recording never raises, and a retrieval that fails leaves the specialist without context rather than failing the turn. Seven collections back the system:\n\n- **BUILTIN** — curated MBA-style domain knowledge shipped with the repo, including a `sales` domain (founder-led sales, qualification and pipeline, proposals and pricing conversations, follow-up and forecasting) and solo-founder guides in finance, marketing, legal and HR. Seeded at boot by `loader.seed_builtin_knowledge` (FAILURES by `seed_failures`): the whole tree on an empty collection, otherwise only the `SHIPPED_BUILTIN_FILES` entries with no chunks yet — so a doc added in a later release reaches an install seeded by an earlier one, instead of shipping, registering as approved, and never being retrievable. Presence is judged from each row's `source` path relative to `knowledge/builtin/` (not the absolute path, which moves with the install), read in one metadata scan narrowed to seeded rows so the external OER corpus sharing the collection is never pulled. A complete store indexes nothing.\n- **COMPANY** — customer documents ingested via the `/documents` endpoint. Curated uploads only — synced Notion pages and Drive files do not land here. A document's identity is its **filename**: chunk metadata and the chunk id both derive from it, so re-uploading upserts in place rather than duplicating, and `DELETE /documents/{filename}` matches the rows it claims to remove. Inbound attachments are **not** here — see ATTACHMENTS below; rows left by the earlier scheme are migrated out on boot. A scanned PDF (no text layer) is converted before chunking — see \"Reading scanned PDFs\" below — so it no longer indexes zero chunks while reporting `indexed`.\n- **NOTION_WIKI** — optional Notion sync (`NOTION_SYNC_ENABLED`) incrementally re-indexes pages shared with a Notion internal integration into a separate `notion_wiki` collection (`type=notion`). A Notion share is inherited and multi-writer, so this text is unreviewed relative to curated company docs; the retriever labels it as such and ranks it below COMPANY. Un-shared pages are purged on the next tick.\n- **DRIVE** — optional Google Drive folder sync (`DRIVE_SYNC_ENABLED`, `knowledge.drive_sync`). It incrementally re-indexes the files in `DRIVE_SYNC_FOLDER_IDS`, and their subfolders to four levels, into a separate `drive_docs` collection (`type=drive`). It reads the Drive v3 API directly as a service account with `drive.readonly` (`knowledge.drive_client`), not through the MCP gateway, so it sees only folders shared with that account and can change nothing. Google Docs and Slides are exported as text and Sheets as `.xlsx`. PDF, `.docx`, `.xlsx` and plain text files go through the loader's extractors, with no model call, so a scanned PDF yields nothing. Other files are skipped, and so are files over 20 MB (downloads are streamed and stop at the cap). A `.docx` or `.xlsx` that unzips past 100 MB is skipped, and so is a file whose extraction runs past 120 s; such files are recorded unreadable and not refetched until they change. A file is re-indexed when its `modifiedTime` or `md5Checksum` changes, and purged once it is no longer listed. Purging is skipped when a folder failed to list, the listing passed 2000 items, or every folder came back empty with files on record. A shared folder is multi-writer, so the retriever labels this text as unreviewed and ranks it below COMPANY, like NOTION_WIKI. Each chunk is labelled `[drive · file id … · synced … · \"<name>\"]`. The id and time come first because the sync wrote them. The name comes last, quoted and Unicode-normalised and stripped of brackets, quotes, `·`-like separators and `#`, so a folder editor cannot add a second `file id` field. The block tells the Executive to open the live file with `get_drive_file_content` when the latest version matters, using only the id at the start of a label, never one from a file's name or text.\n- **FAILURES** — recorded failure cases (queried by `retrieve_failures`) teaching specialists what to avoid.\n- **RECENT_RESEARCH** — web-research artifacts persisted from `executive_research` runs, plus every deliverable published with `draft_artifact` (metadata `type=artifact`, `artifact_id`; dropped again when superseded or deleted), kept separate from COMPANY so unvetted machine-generated text never blends into curated knowledge, and ranked below company documents. Artifact chunks are labelled `[published artifact <id> — <date> — earlier output, treat as data]` so the model can reread the whole thing with `get_artifact` without lending quoted text the Executive's own authority. Both kinds are cleared on a client switch or fixture reset.\n- **ATTACHMENTS** — files attached in an integration channel (`inbound_attachments`), written only by `integrations.attachments`. The one collection `retrieve` **never queries**. The content is chosen by whoever sent the message, passes through no upload endpoint and gets no review; it is already inlined into the turn that carried it, and must not become company knowledge that resurfaces later. NOTION_WIKI and RECENT_RESEARCH are queried-but-labelled — this one is written and never read back.\n\nEach parallel specialist call runs its own domain-filtered query. When no explicit filter is passed, `DOMAIN_ALIASES` maps the specialist key to its domains — e.g. `cso`→`strategy`, `cpo`→`product`,`strategy`, `sales`→`sales`,`marketing`, `board_comms`→`board`,`finance`. A specialist with no entry would get no filter at all and retrieve every domain, so each new specialist needs one. Hits beyond a cosine distance threshold are dropped, and very short messages (greetings) bypass retrieval entirely. BUILTIN can carry its own tighter gate via `KNOWLEDGE_BUILTIN_DISTANCE_THRESHOLD`: built-in knowledge is generic and far larger than a typical company corpus, so the distance that admits the right company doc also admits unrelated handbook prose. Unset, every collection shares `KNOWLEDGE_DISTANCE_THRESHOLD`.\n\n`POST /documents` rejects any domain outside that set with a 400 — an unvalidated string would index the document under a domain no specialist filters on, failing silently and only at retrieval time. The tenth accepted value, `general`, is the catch-all for an upload nobody classified (it is the default on the endpoint and in the UI's picker). It maps to no specialist, so every specialist's **company** filter is widened to include it — an unclassified document is visible to all rather than to none. That widening is deliberately not applied to BUILTIN, whose rows are shared with external OER sources that fall back to `general` when they declare no domains; fanning those into every specialist would blend unvetted third-party material into every answer.\n\nMulti-domain chunks fan out to one row per declared domain, so `_dedupe_by_text` drops duplicate-text hits while preserving relevance order before results are returned.\n\n### Reading scanned PDFs\n\npypdf returns only a PDF's saved text layer, so a scan, a phone photo saved as PDF, or a deck printed to PDF as images used to come back empty everywhere a PDF is read. `knowledge.pdf_reader.read_pdf_text` tries three readers in order and never raises:\n\n1. **Text layer** (pypdf) — used whenever it yields real text (at least ~20 visible characters per page, so a scan carrying only page numbers still falls through).\n2. **The deployment's own model, through its provider's PDF support** — opt-in, `PDF_PROVIDER_READING=true` (off by default: an upgrade never starts sending scanned PDFs off the server; this also means an Anthropic-direct deployment OCRs scans locally until it opts in) — the pages as a base64 `document` block, transcribed to Markdown by `PDF_VISION_MODEL` (unset: `DEFAULT_MODEL`, the model the deployment already runs on), `PDF_VISION_PAGES_PER_CALL` pages per request (default 20), three requests at a time. Each provider carries the block its own way: Anthropic reads it natively; on OpenRouter the translator makes it an OpenAI `file` part and the provider adds the `file-parser` plugin — engine `native` for Claude or any model the catalog lists with `file` input (OpenAI's GPT-4o and later, Gemini…), else `PDF_OPENROUTER_ENGINE` (default `mistral-ocr`, $2 per 1,000 pages); a local OpenAI-compatible server gets the `file` part only with `LOCAL_PDF_INPUT=true` (e.g. `LOCAL_BASE_URL` pointed at OpenAI's own API). `providers.registry.pdf_input_supported` decides; a model that cannot take one never receives it (the feature gate swaps a stray `document` block for a note). Turning it on sends scanned pages off the server — to the provider, and on OpenRouter possibly to its parser (Mistral OCR) — and the destination is logged once per process.\n3. **Local OCR** — pages rendered with pypdfium2 and read by RapidOCR (ONNX, models bundled in the wheel): offline, no key, no per-page cost. The default reader while step 2 is off, what a local model without PDF input gets, and the fallback on every provider when step 2 errors, refuses or returns nothing. `PDF_OCR_ENABLED=false` turns it off.\n\n`PDF_VISION_MAX_PAGES` (default 100) caps the pages read, with a note saying how many were; a file a channel delivered (`inbound=True`) is capped at `PDF_INBOUND_MAX_PAGES` (30) and metered by a rolling `PDF_INBOUND_PAGES_PER_HOUR` (300) budget. OCR is bounded too: a page renders within a 25-megapixel budget whatever its MediaBox claims (a crafted 20000pt page would otherwise allocate gigabytes), at most two documents OCR at once, and each stops after 240 s with what it has. A PDF of more than 2,000 pages is refused before any page is parsed (the caps above bound conversion, not the text-layer pass). A Claude slice that runs into the output limit is split in half until each part fits; a single page that still does not fit keeps what was written, marked `[transcription cut off here]`, and the note says so. A result is cached in process by the file's sha256 (32 entries) so a re-read does not re-bill; a failure is not cached. `loader.read_document_text` / `extract_text_from_file_async` route a PDF through it and every other type through the sync extractor, and are what `ingest_file` (uploads, boot reconcile), the `/documents/{filename}` preview, intake uploads, chat attachments, the email poller, `oe__read_file` and the Executive's `read_document` tool use.\n\n### The SME review gate\n\nBefore the distance threshold applies, both `retrieve` and `retrieve_failures` drop **withheld** items — anything the review store reports as `pending` or `rejected` (`ReviewStore.get_withheld_filenames`). Approved items with a priority set are then sorted ahead of the rest.\n\nKnowledge shipped with the product registers as a trusted default (`approved`, NULL `reviewed_at`), so nothing is withheld until a human deliberately queues a domain for curation or rejects something. See the **SME Knowledge Review** section for the full state model. The filter matches on the `(domain, filename)` pair Chroma carries. A bare filename was ambiguous *and* caller-choosable — `POST /knowledge/builtin` takes any `(domain, filename)`, so uploading a file named after a shipped doc in another domain silently withheld that shipped doc from every specialist. Failure case studies are keyed in their own namespace, so a user-authored one cannot collide with a shipped one.\n\nThe formatted context is injected into the **user turn** (wrapped in `<relevant_knowledge>` by `BaseAgent.analyze`), never into a cached system block — per-query hits would otherwise poison the prompt cache.\n\n### Skills (Playbooks) collection\n\nSkills — called **playbooks** in the UI, where they are a tab on Workflows — are Markdown procedures with YAML frontmatter, indexed one document per skill in their own `skills` collection (name + description + when_to_use; the body is fetched by `load_skill`). Built-ins ship under `knowledge/builtin/skills/`; company skills live in `company/skills/`. Customizing a built-in (a `PUT` from the Playbooks tab; the chat tools refuse built-ins) writes a company skill of the same name flagged `customizes_builtin: true`, which **replaces** the built-in: it shadows the built-in everywhere (chat, and the workflows that follow it — see the Workflows section), and deleting it reverts. A built-in cannot be deleted, so \"delete\" **hides** it via `company/skills/.hidden.yaml`; hidden and shadowed built-ins are dropped from the index. `sync_builtin_skill_index` reconciles the built-in rows at startup, after a client-slot restore, and after a fixture reset — so a built-in added in a release reaches an already-seeded index, and a removed one leaves it.\n\n**Drafts.** The Executive's chat tools never change the library directly: `create_skill`, `update_skill` and `delete_skill` save a draft as JSON in `company/skills/.drafts/` (one per name; a newer proposal replaces the pending one). Drafts are not `*.md`, so they are never listed, indexed, searched, loaded or followed. A person approves a draft on the Playbooks tab (`POST /skill-drafts/{name}/approve`), which applies it through the ordinary skills operations, or discards it. Each save gets a fresh `id`, and approve/discard name the version the person saw: a proposal swapped in after review (say by a crafted inbound message) is never applied unseen, and approving never removes a newer draft that arrived meanwhile. This matches the draft → human-approval shape of custom workflows: a crafted inbound message can propose a playbook, but only a person can put one in effect.",+ "markdown": "Retrieval is handled by `openexecutive.knowledge.retriever.retrieve`, querying ChromaDB collections through `openexecutive.knowledge.store.ChromaDBStore`. When the caller passes `record_source`, `retrieve` also names every document its block quotes — company uploads by filename, built-in knowledge by a title made from its filename, Notion pages, synced Drive files by name with their Drive link, research notes by date, and earlier published documents with a link to `/artifacts/<id>` — so the web chat can list what an answer looked at: the chat route's own search, and each specialist's that answered (see the Lifecycle and API sections). The recording never raises, and a retrieval that fails leaves the specialist without context rather than failing the turn. Seven collections back the system:\n\n- **BUILTIN** — curated MBA-style domain knowledge shipped with the repo, including a `sales` domain (founder-led sales, qualification and pipeline, proposals and pricing conversations, follow-up and forecasting) and solo-founder guides in finance, marketing, legal and HR. Seeded at boot by `loader.seed_builtin_knowledge` (FAILURES by `seed_failures`): the whole tree on an empty collection, otherwise only the `SHIPPED_BUILTIN_FILES` entries with no chunks yet — so a doc added in a later release reaches an install seeded by an earlier one, instead of shipping, registering as approved, and never being retrievable. Presence is judged from each row's `source` path relative to `knowledge/builtin/` (not the absolute path, which moves with the install), read in one metadata scan narrowed to seeded rows so the external OER corpus sharing the collection is never pulled. A complete store indexes nothing.\n- **COMPANY** — customer documents ingested via the `/documents` endpoint. Curated uploads only — synced Notion pages and Drive files do not land here. A document's identity is its **filename**: chunk metadata and the chunk id both derive from it, so re-uploading upserts in place rather than duplicating, and `DELETE /documents/{filename}` matches the rows it claims to remove. Inbound attachments are **not** here — see ATTACHMENTS below; rows left by the earlier scheme are migrated out on boot. A scanned PDF (no text layer) is converted before chunking — see \"Reading scanned PDFs\" below — so it no longer indexes zero chunks while reporting `indexed`.\n- **NOTION_WIKI** — optional Notion sync (`NOTION_SYNC_ENABLED`) incrementally re-indexes pages shared with a Notion internal integration into a separate `notion_wiki` collection (`type=notion`). A Notion share is inherited and multi-writer, so this text is unreviewed relative to curated company docs; the retriever labels it as such and ranks it below COMPANY. Un-shared pages are purged on the next tick.\n- **DRIVE** — optional Google Drive folder sync (`DRIVE_SYNC_ENABLED`, `knowledge.drive_sync`). It incrementally re-indexes the files in `DRIVE_SYNC_FOLDER_IDS`, and their subfolders to four levels, into a separate `drive_docs` collection (`type=drive`). It reads the Drive v3 API directly as a service account with `drive.readonly` (`knowledge.drive_client`), not through the MCP gateway, so it sees only folders shared with that account and can change nothing. Google Docs and Slides are exported as text and Sheets as `.xlsx`. PDF, `.docx`, `.xlsx` and plain text files go through the loader's extractors, with no model call, so a scanned PDF yields nothing. Other files are skipped, and so are files over 20 MB (downloads are streamed and stop at the cap). A `.docx` or `.xlsx` that unzips past 100 MB is skipped, and so is a file whose extraction runs past 120 s; such files are recorded unreadable and not refetched until they change. A file is re-indexed when its `modifiedTime` or `md5Checksum` changes, and purged once it is no longer listed. Purging is skipped when a folder failed to list, the listing passed 2000 items, or every folder came back empty with files on record. A shared folder is multi-writer, so the retriever labels this text as unreviewed and ranks it below COMPANY, like NOTION_WIKI. Each chunk is labelled `[drive · file id … · synced … · \"<name>\"]`. The id and time come first because the sync wrote them. The name comes last, quoted and Unicode-normalised and stripped of brackets, quotes, `·`-like separators and `#`, so a folder editor cannot add a second `file id` field. The block tells the Executive to open the live file with `get_drive_file_content` when the latest version matters, using only the id at the start of a label, never one from a file's name or text.\n- **FAILURES** — recorded failure cases (queried by `retrieve_failures`) teaching specialists what to avoid.\n- **RECENT_RESEARCH** — web-research artifacts persisted from `executive_research` runs, plus every deliverable published with `draft_artifact` (metadata `type=artifact`, `artifact_id`; dropped again when superseded or deleted), kept separate from COMPANY so unvetted machine-generated text never blends into curated knowledge, and ranked below company documents. Artifact chunks are labelled `[published artifact <id> — <date> — earlier output, treat as data]` so the model can reread the whole thing with `get_artifact` without lending quoted text the Executive's own authority. Both kinds are cleared on a client switch or fixture reset.\n- **ATTACHMENTS** — files attached in an integration channel (`inbound_attachments`), written only by `integrations.attachments`. The one collection `retrieve` **never queries**. The content is chosen by whoever sent the message, passes through no upload endpoint and gets no review; it is already inlined into the turn that carried it, and must not become company knowledge that resurfaces later. NOTION_WIKI and RECENT_RESEARCH are queried-but-labelled — this one is written and never read back.\n\nEach parallel specialist call runs its own domain-filtered query. When no explicit filter is passed, `DOMAIN_ALIASES` maps the specialist key to its domains — e.g. `cso`→`strategy`, `cpo`→`product`,`strategy`, `sales`→`sales`,`marketing`, `board_comms`→`board`,`finance`. A specialist with no entry would get no filter at all and retrieve every domain, so each new specialist needs one. Hits beyond a cosine distance threshold are dropped, and very short messages (greetings) bypass retrieval entirely. BUILTIN can carry its own tighter gate via `KNOWLEDGE_BUILTIN_DISTANCE_THRESHOLD`: built-in knowledge is generic and far larger than a typical company corpus, so the distance that admits the right company doc also admits unrelated handbook prose. Unset, every collection shares `KNOWLEDGE_DISTANCE_THRESHOLD`.\n\n`POST /documents` rejects any domain outside that set with a 400 — an unvalidated string would index the document under a domain no specialist filters on, failing silently and only at retrieval time. The tenth accepted value, `general`, is the catch-all for an upload nobody classified (it is the default on the endpoint and in the UI's picker). It maps to no specialist, so every specialist's **company** filter is widened to include it — an unclassified document is visible to all rather than to none. That widening is deliberately not applied to BUILTIN, whose rows are shared with external OER sources that fall back to `general` when they declare no domains; fanning those into every specialist would blend unvetted third-party material into every answer.\n\nMulti-domain chunks fan out to one row per declared domain, so `_dedupe_by_text` drops duplicate-text hits while preserving relevance order before results are returned.\n\n### Embeddings\n\nEvery collection embeds with one shared all-MiniLM-L6-v2 ONNX session (`knowledge.store._embedding_function`), the same model and vectors as Chroma's default. Chroma's own default reloads the model on every call and keeps ONNX Runtime's memory arena, so a 300-chunk upload peaked near 760 MB and each query spent ~240 ms reloading. The shared session is built once, runs without the arena and embeds four texts at a time (the model pads every text to 256 tokens, so batch size sets the peak): ~300 MB and ~30 ms. It answers to Chroma's name `default` with an empty config, because every existing collection records that name and Chroma refuses a collection opened with a differently named function; and it must not be a `DefaultEmbeddingFunction` instance, which Chroma swaps for a fresh default on every embed.\n\n### Reading scanned PDFs\n\npypdf returns only a PDF's saved text layer, so a scan, a phone photo saved as PDF, or a deck printed to PDF as images used to come back empty everywhere a PDF is read. `knowledge.pdf_reader.read_pdf_text` tries three readers in order and never raises:\n\n1. **Text layer** (pypdf) — used whenever it yields real text (at least ~20 visible characters per page, so a scan carrying only page numbers still falls through).\n2. **The deployment's own model, through its provider's PDF support** — opt-in, `PDF_PROVIDER_READING=true` (off by default: an upgrade never starts sending scanned PDFs off the server; this also means an Anthropic-direct deployment OCRs scans locally until it opts in) — the pages as a base64 `document` block, transcribed to Markdown by `PDF_VISION_MODEL` (unset: `DEFAULT_MODEL`, the model the deployment already runs on), `PDF_VISION_PAGES_PER_CALL` pages per request (default 20), three requests at a time. Each provider carries the block its own way: Anthropic reads it natively; on OpenRouter the translator makes it an OpenAI `file` part and the provider adds the `file-parser` plugin — engine `native` for Claude or any model the catalog lists with `file` input (OpenAI's GPT-4o and later, Gemini…), else `PDF_OPENROUTER_ENGINE` (default `mistral-ocr`, $2 per 1,000 pages); a local OpenAI-compatible server gets the `file` part only with `LOCAL_PDF_INPUT=true` (e.g. `LOCAL_BASE_URL` pointed at OpenAI's own API). `providers.registry.pdf_input_supported` decides; a model that cannot take one never receives it (the feature gate swaps a stray `document` block for a note). Turning it on sends scanned pages off the server — to the provider, and on OpenRouter possibly to its parser (Mistral OCR) — and the destination is logged once per process.\n3. **Local OCR** — pages rendered with pypdfium2 and read by RapidOCR (ONNX, models bundled in the wheel): offline, no key, no per-page cost. The default reader while step 2 is off, what a local model without PDF input gets, and the fallback on every provider when step 2 errors, refuses or returns nothing. `PDF_OCR_ENABLED=false` turns it off.\n\n`PDF_VISION_MAX_PAGES` (default 100) caps the pages read, with a note saying how many were; a file a channel delivered (`inbound=True`) is capped at `PDF_INBOUND_MAX_PAGES` (30) and metered by a rolling `PDF_INBOUND_PAGES_PER_HOUR` (300) budget. OCR is bounded too: a page renders within a 25-megapixel budget whatever its MediaBox claims (a crafted 20000pt page would otherwise allocate gigabytes), at most two documents OCR at once, and each stops after 240 s with what it has. A PDF of more than 2,000 pages is refused before any page is parsed (the caps above bound conversion, not the text-layer pass). A Claude slice that runs into the output limit is split in half until each part fits; a single page that still does not fit keeps what was written, marked `[transcription cut off here]`, and the note says so. A result is cached in process by the file's sha256 (32 entries) so a re-read does not re-bill; a failure is not cached. `loader.read_document_text` / `extract_text_from_file_async` route a PDF through it and every other type through the sync extractor, and are what `ingest_file` (uploads, boot reconcile), the `/documents/{filename}` preview, intake uploads, chat attachments, the email poller, `oe__read_file` and the Executive's `read_document` tool use.\n\n### The SME review gate\n\nBefore the distance threshold applies, both `retrieve` and `retrieve_failures` drop **withheld** items — anything the review store reports as `pending` or `rejected` (`ReviewStore.get_withheld_filenames`). Approved items with a priority set are then sorted ahead of the rest.\n\nKnowledge shipped with the product registers as a trusted default (`approved`, NULL `reviewed_at`), so nothing is withheld until a human deliberately queues a domain for curation or rejects something. See the **SME Knowledge Review** section for the full state model. The filter matches on the `(domain, filename)` pair Chroma carries. A bare filename was ambiguous *and* caller-choosable — `POST /knowledge/builtin` takes any `(domain, filename)`, so uploading a file named after a shipped doc in another domain silently withheld that shipped doc from every specialist. Failure case studies are keyed in their own namespace, so a user-authored one cannot collide with a shipped one.\n\nThe formatted context is injected into the **user turn** (wrapped in `<relevant_knowledge>` by `BaseAgent.analyze`), never into a cached system block — per-query hits would otherwise poison the prompt cache.\n\n### Skills (Playbooks) collection\n\nSkills — called **playbooks** in the UI, where they are a tab on Workflows — are Markdown procedures with YAML frontmatter, indexed one document per skill in their own `skills` collection (name + description + when_to_use; the body is fetched by `load_skill`). Built-ins ship under `knowledge/builtin/skills/`; company skills live in `company/skills/`. Customizing a built-in (a `PUT` from the Playbooks tab; the chat tools refuse built-ins) writes a company skill of the same name flagged `customizes_builtin: true`, which **replaces** the built-in: it shadows the built-in everywhere (chat, and the workflows that follow it — see the Workflows section), and deleting it reverts. A built-in cannot be deleted, so \"delete\" **hides** it via `company/skills/.hidden.yaml`; hidden and shadowed built-ins are dropped from the index. `sync_builtin_skill_index` reconciles the built-in rows at startup, after a client-slot restore, and after a fixture reset — so a built-in added in a release reaches an already-seeded index, and a removed one leaves it.\n\n**Drafts.** The Executive's chat tools never change the library directly: `create_skill`, `update_skill` and `delete_skill` save a draft as JSON in `company/skills/.drafts/` (one per name; a newer proposal replaces the pending one). Drafts are not `*.md`, so they are never listed, indexed, searched, loaded or followed. A person approves a draft on the Playbooks tab (`POST /skill-drafts/{name}/approve`), which applies it through the ordinary skills operations, or discards it. Each save gets a fresh `id`, and approve/discard name the version the person saw: a proposal swapped in after review (say by a crafted inbound message) is never applied unseen, and approving never removes a newer draft that arrived meanwhile. This matches the draft → human-approval shape of custom workflows: a crafted inbound message can propose a playbook, but only a person can put one in effect.","mermaid": "flowchart TD\n Call[\"Specialist call<br/>(domain-filtered)\"] --> Q[\"retrieve(query, domain)\"]\n Q --> Builtin[(BUILTIN)]\n Q --> Company[(COMPANY)]\n Q --> NotionWiki[(NOTION_WIKI)]\n Q --> DriveDocs[(DRIVE)]\n Q --> Failures[(FAILURES)]\n Q --> Research[(RECENT_RESEARCH)]\n Attach[\"integrations.attachments\"] --> Attachments[(ATTACHMENTS<br/>never queried)]\n Builtin --> Dedupe[\"dedupe by text<br/>+ distance gate\"]\n Company --> Dedupe\n NotionWiki --> Dedupe\n DriveDocs --> Dedupe\n Failures --> Dedupe\n Research --> Dedupe\n Dedupe --> Inject[\"Inject into USER turn<br/>(not cached)\"]\n class Call entry\n class Q compute\n class Dedupe compute\n class Builtin storage\n class Company storage\n class NotionWiki storage\n class DriveDocs storage\n class Failures storage\n class Research storage\n class Attachments storage\n class Attach compute\n class Inject hot",- "generated_at": "2026-09-29T22:37:08.352837+00:00"+ "generated_at": "2026-09-30T15:22:47.773079+00:00"}diff --git a/packages/core/openexecutive/knowledge/store.py b/packages/core/openexecutive/knowledge/store.pyindex d825addf..bf66e91d 100644--- a/packages/core/openexecutive/knowledge/store.py+++ b/packages/core/openexecutive/knowledge/store.py@@ -1,10 +1,78 @@from __future__ import annotationsimport logging+import os+import threadingfrom abc import ABC, abstractmethodfrom pathlib import Pathfrom typing import Any+# Chroma's DefaultEmbeddingFunction builds a new ONNXMiniLM_L6_V2 on every+# call, so every query and upsert loads the model from disk (about 200 ms a+# query). Its ONNX session also keeps the CPU memory arena on and embeds 32+# texts per run, each padded to 256 tokens, so a large ingest peaks several+# hundred MB above the model. The stores share one session per process+# instead, with the arena off and _EMBED_BATCH texts per run: the same model,+# the same vectors, a bounded peak.+#+# It answers to DefaultEmbeddingFunction's name ("default") with an empty+# config, which is what existing collections record: Chroma refuses to open+# a collection with a differently named function, and rebuilds "default"+# from the stored config if older code opens one this created. It must NOT+# be a DefaultEmbeddingFunction instance: Chroma's Collection._embed skips+# any such instance and embeds with a fresh one from the config instead.+_EMBED_BATCH = 4+_embedding_function_lock = threading.Lock()+_shared_embedding_function: Any = None+++def _embedding_function() -> Any:+ global _shared_embedding_function+ with _embedding_function_lock:+ if _shared_embedding_function is None:+ _shared_embedding_function = _build_embedding_function()+ return _shared_embedding_function+++def _build_embedding_function() -> Any:+ from chromadb.utils.embedding_functions.onnx_mini_lm_l6_v2 import ONNXMiniLM_L6_V2++ class _SharedMiniLM(ONNXMiniLM_L6_V2): # type: ignore[misc]+ _session: Any = None+ _session_lock = threading.Lock()++ @staticmethod+ def name() -> str:+ return "default"++ def get_config(self) -> dict[str, Any]:+ return {}++ @property+ def model(self) -> Any:+ with self._session_lock:+ if self._session is None:+ so = self.ort.SessionOptions()+ so.log_severity_level = 3+ so.graph_optimization_level = self.ort.GraphOptimizationLevel.ORT_ENABLE_ALL+ so.enable_cpu_mem_arena = False+ # Chroma's own choice: every available provider but CoreML.+ providers = [+ p for p in self.ort.get_available_providers()+ if p != "CoreMLExecutionProvider"+ ]+ self._session = self.ort.InferenceSession(+ os.path.join(self.DOWNLOAD_PATH, self.EXTRACTED_FOLDER_NAME, "model.onnx"),+ providers=providers,+ sess_options=so,+ )+ return self._session++ def _forward(self, documents: list[str], batch_size: int = _EMBED_BATCH) -> Any:+ return super()._forward(documents, batch_size=_EMBED_BATCH)++ return _SharedMiniLM()+class KnowledgeStore(ABC):@abstractmethod@@ -84,6 +152,7 @@ class ChromaDBStore(KnowledgeStore):return self._client.get_or_create_collection(name=name,metadata={"hnsw:space": "cosine"},+ embedding_function=_embedding_function(),)def add_documents(diff --git a/packages/core/tests/unit/test_knowledge_store_embedding.py b/packages/core/tests/unit/test_knowledge_store_embedding.pynew file mode 100644index 00000000..409f09a9--- /dev/null+++ b/packages/core/tests/unit/test_knowledge_store_embedding.py@@ -0,0 +1,99 @@+"""The embedding function every ChromaDBStore collection uses.++Chroma's DefaultEmbeddingFunction reloads the model on every call and keeps+the ONNX memory arena, so the store shares one lean session instead. Existing+collections record the function they were created with, so the shared one must+answer to the same name and produce the same vectors.+"""++from __future__ import annotations++from pathlib import Path++import chromadb+import numpy as np+from chromadb.config import Settings+from chromadb.utils.embedding_functions import DefaultEmbeddingFunction++from openexecutive.knowledge import store as store_mod+from openexecutive.knowledge.store import ChromaDBStore+++def test_one_function_is_shared_under_chromas_default_name() -> None:+ ef = store_mod._embedding_function()+ assert ef is store_mod._embedding_function()+ assert ef.name() == DefaultEmbeddingFunction.name() == "default"+ assert ef.get_config() == DefaultEmbeddingFunction().get_config()+ # Chroma's Collection._embed ignores a DefaultEmbeddingFunction instance+ # and embeds with a fresh one rebuilt from the collection config.+ assert not isinstance(ef, DefaultEmbeddingFunction)+++def test_vectors_match_chromas_default() -> None:+ texts = ["quarterly revenue forecast for the board", "hiring plan", "word " * 300]+ ours = np.array(store_mod._embedding_function()(texts))+ chromas = np.array(DefaultEmbeddingFunction()(texts))+ assert np.allclose(ours, chromas, atol=1e-5)+++def test_session_runs_without_the_memory_arena() -> None:+ ef = store_mod._embedding_function()+ ef(["warm"])+ session = ef.model+ assert session.get_session_options().enable_cpu_mem_arena is False+ assert session is ef.model # built once, not per call+++def test_embeds_a_few_texts_per_run(monkeypatch) -> None:+ minilm = store_mod._embedding_function()+ runs: list[int] = []+ real_run = minilm.model.run++ class Spy:+ def run(self, names, feed):+ runs.append(len(feed["input_ids"]))+ return real_run(names, feed)++ monkeypatch.setattr(minilm, "_session", Spy())+ minilm(["text"] * 10)+ assert runs == [store_mod._EMBED_BATCH] * 2 + [10 - 2 * store_mod._EMBED_BATCH]+++def test_opens_a_collection_created_with_chromas_default(tmp_path: Path) -> None:+ # Every collection on an existing install was created this way.+ client = chromadb.PersistentClient(path=str(tmp_path), settings=Settings(anonymized_telemetry=False))+ col = client.get_or_create_collection(ChromaDBStore.COMPANY_COLLECTION, metadata={"hnsw:space": "cosine"})+ col.upsert(+ documents=["quarterly revenue forecast for the board", "engineering hiring plan"],+ metadatas=[{"domain": "finance"}, {"domain": "operations"}],+ ids=["revenue", "hiring"],+ )++ store = ChromaDBStore(persist_directory=tmp_path)+ hits = store.query("revenue forecast", collection=ChromaDBStore.COMPANY_COLLECTION, n_results=1)+ assert hits[0]["metadata"]["domain"] == "finance"+ store.add_documents(+ ["vendor contract renewal terms"], [{"domain": "legal"}], ["contract"],+ collection=ChromaDBStore.COMPANY_COLLECTION,+ )+ assert store.get_collection_count(ChromaDBStore.COMPANY_COLLECTION) == 3+++def test_store_queries_and_upserts_embed_through_the_shared_session(tmp_path: Path, monkeypatch) -> None:+ ef = store_mod._embedding_function()+ batches: list[int] = []+ real_forward = type(ef)._forward++ def spy(self, documents, batch_size=store_mod._EMBED_BATCH):+ batches.append(len(documents))+ return real_forward(self, documents, batch_size)++ monkeypatch.setattr(type(ef), "_forward", spy)+ store = ChromaDBStore(persist_directory=tmp_path)+ store.add_documents(+ ["quarterly revenue forecast", "engineering hiring plan"],+ [{"domain": "finance"}, {"domain": "operations"}], ["a", "b"],+ collection=ChromaDBStore.COMPANY_COLLECTION,+ )+ store.query("revenue", collection=ChromaDBStore.COMPANY_COLLECTION, n_results=1)+ assert batches == [2, 1]