Decile Hub Connector

Decile Hub Connector

Decile Hub is a VC fund-management + CRM platform. The Decile Hub API v1 is the first per-client custom connector in the Lossless tree: each VC client has its own Decile tenant (subdomain), its own API token, and its own clients/<slug>/.env. This skill is the operating guide for pulling from and pushing to that API, and for mapping its records into our SurrealDB canonical layer.

Source of truth. The authoritative contract is the on-disk OpenAPI 3.0.1 spec: ai-labs/augment-it/clients/humain-vc/inputs/decilehub/202506_decilehub-docs_swagger.yaml (11,970 lines). The full endpoint inventory lives in references/endpoint-inventory.md. When in doubt, read the spec — do not paraphrase Decile's API from memory.

When to use this skill

• Pulling data from Decile (list/get people, organizations, pipeline prospects, deals, funds, portfolio companies, …)
• Pushing data to Decile (create/upsert people & organizations, add prospects, append notes, create tasks, …)
• Wiring the Decile connector for a new client (new tenant subdomain + token in that client's .env)
• Building or maintaining the decile-mcp server (ai-labs/augment-it/services/decile-mcp/)
• Reconciling Decile records into SurrealDB persons / organizations

Connection contract

Env vars (live in the per-client .env)

Decile is tenant-scoped, so its config belongs in clients/<slug>/.env, resolved through the workspace connector seam (services/workspace/) — not in a shared root .env.

DECILE_API_URL=https://humain.decilehub.com      # the tenant's base URL
DECILE_HUB_API_KEY=<the API token from /settings/api>   # sent raw as the Authorization header

These are Decile's own naming. The earlier spec/README anticipated DECILE_API_BASE_URL / DECILE_API_KEY / DECILE_TENANT_ID; we standardize on the real names above and the tenant is encoded in the URL (no separate tenant id needed).

The canonical request shape

const res = await fetch(`${DECILE_API_URL}/api/v1/whoami`, {
  headers: { Authorization: DECILE_HUB_API_KEY, Accept: 'application/json' },
});

Pulling data (reads)

Reads are GET /api/v1/<resource> (list) and GET /api/v1/<resource>/{id} (show). Two cross-cutting concerns:

⚠️ There are THREE pagination patterns — do not assume one

The API is not uniform. Detect the pattern per endpoint group (see the inventory for which is which):

Filtering & custom data points

• Most list endpoints accept resource-specific filters (name, email, created_after, stage_name, …) — see the inventory.
• custom_data_points query param on people/organizations/pipeline_prospects list+show: * = all, comma-list = subset, empty = none. Select-type values resolve to human-readable labels on read; internal jsonb keys are never returned.
• include pulls associations (notes, people, organizations, referred_by, …); fields narrows the response.

Pushing data (writes)

Prefer the upsert endpoints — they're idempotent and map cleanly to our model

The singular upsert routes (/person, /organization, /pipeline_prospect — note: singular) match-or-create by natural key and return a changes diff. This is the right default for sync.

Bulk create = dedup, not upsert

POST /api/v1/people, /organizations, /pipeline_prospects (plural) process the first 100 and return { created, duplicates, errors }. Duplicates (by email / name) are skipped, not updated — use these for first-load, the singular upserts for ongoing sync.

Other common writes

• Notes: POST /api/v1/{people|organizations}/{id}/notes and /pipeline_prospects/{id}/notes — body { note: { body, context } }.
• Tags: tag_list (comma-separated string) adds; remove_tag_list removes (upsert routes only).
• Custom data points (write): the custom_data_points object in person/org/prospect bodies. New fields are defined via POST /api/v1/pipelines/{pipeline_id}/data_points (account admin; format enum incl. string, select, currency_us, url, …).
• Not idempotent: POST /entities and journal-entry creates re-create on retry — GET first to check.

Write fields — people & organizations

There is no standalone Person/Organization schema — stored fields are dynamic (data / custom_data_points jsonb). The documented write fields:

• Person: first_name, last_name, email*, middle_name, phone, linkedin, tag_list, custom_data_points, note, picture (base64/URL), address, referred_by, organizations: [{ name, title }].
• Organization: name*, website, description, tag_list, logo, custom_data_points, note, address, referred_by, people: [associated_person].

Files and attachments — three surfaces, and they are not interchangeable

Absent from the June 2026 swagger snapshot. The /pipeline_prospects/{id}/attachments family exists only in the live docs at https://<tenant>.decilehub.com/docs/api — the on-disk spec has just /api/v1/base/attachments/{id}. Verified live 2026-08-19; treat live docs as authoritative where the two disagree.

POST /api/v1/pipeline_prospects/{pipeline_prospect_id}/attachments

multipart/form-data — attachment[file] (binary, required) and optional attachment[name] (defaults to the filename without its extension, matching the web UI). No file-type allow-list is applied here — web-UI parity; the data room upload endpoint is stricter.

• investment-type pipelines only — a prospect on a closing/investor pipeline returns 403.
• Requires pipeline edit access (edit_prospects?), not just read.
• 201 returns the same entry shape as the list endpoint, with id and signed_id.
• 400 missing attachment[file] · 403 wrong pipeline type or no edit access · 404 prospect not in the caller's account · 422 validation.

House naming convention for decks: <date>_<CompanyName>--<Round>.pdf (202608_ImpulseLabs--Pre-Seed.pdf) — YYYYMMDD when the send date is known, YYYYMM when only the month is. VCs see the same company at multiple rounds; the round token is what keeps a re-pitch distinguishable from the original. Full rule in the decilehub-interface skill.

The decile-mcp tool upload_prospect_attachment fronts this with file_data_base64 instead of multipart — the server does the conversion. Don't infer the REST contract from the MCP tool's shape.

GET .../attachments merges two sources — know which you're holding

Entry shape: signed_id, filename, content_type, byte_size, source, name, item_id, uploaded_at.

• source: "direct" — CRM attachments on the org/person, from the Hub UI or from POST /files with attachable_type/attachable_id. Carry a name.
• source: "questionnaire" — files the founder uploaded through a deal-intake questionnaire (e.g. Submit Your Company): pitch decks, cap tables, supporting docs. Carry an item_id matching the filename custom field stored on the organization record.

This is where inbound decks actually live. A company that pitched through the intake form has already delivered its deck — check for a questionnaire entry before asking anyone to send one, or uploading your own copy.

Only direct entries can be deleted (DELETE .../attachments/{attachment_id}); questionnaire uploads are the founder's submission and are not yours to remove. Fetch bytes with GET .../attachments/{signed_id}/download — by signed_id, not id.

Errors

Canonical shape (used on most 4xx):

{ "error": { "code": "validation_failed", "message": "...", "field": null, "valid_values": null, "details": null } }

• Common codes: forbidden, bad_request, not_found, validation_failed, invalid_parameter, confirmation_required, unresolved_variables, already_finalized, …
• Inconsistency to handle: a few endpoints (e.g. single PATCH /pipeline_prospects/{id} on 400/404/422) return a bare { error: "string" } — the client must tolerate both shapes.
• No rate-limit headers and no webhooks are defined in the spec. Async jobs poll a status_url (e.g. financial reports); some actions return 202 (enqueued).

Mapping Decile → SurrealDB canonical layer

Decile is a per-client source; everything written into our canonical layer must carry the client tag (see [[Client-Tagging-on-Canonical-Writes]]). The natural mapping:

Decile's upsert-by-natural-key + changes diff mirrors our own upsert discipline (SELECT-by-key → MERGE/CREATE). When syncing Decile → SurrealDB, treat Decile as one source and record provenance; do not let a Decile refresh overwrite operator-curated commentary. See the SurrealDB connection contract in [[Connecting-To-And-Using-SurrealDB]].

The two surfaces this skill backs

1. This skill — the operating guide (you're reading it).
2. The decile-mcp server — ai-labs/augment-it/services/decile-mcp/ (TypeScript): a typed client that resolves base URL + token from the per-client .env, normalizes the three pagination patterns and the error shape, and exposes Decile operations as MCP tools. The spec marks agent-facing operations with x-agent-tool: true — those are the tools to expose first. Register with claude mcp add -s project.

See also

• references/endpoint-inventory.md — the exhaustive endpoint list, grouped by tag
• The OpenAPI spec: ai-labs/augment-it/clients/humain-vc/inputs/decilehub/202506_decilehub-docs_swagger.yaml
• [[Connecting-To-And-Using-SurrealDB]] — the canonical-layer connection + client-tagging contract
• [[Workspaces-as-Tenant-Primitive]] — the per-client connector seam Decile plugs into
• [[Client-Tagging-on-Canonical-Writes]] — every canonical write carries its client