graphistry-rest-api（graphistry rest API 搜索）

Graphistry REST API

Scope

Use this skill for Graphistry REST endpoint tasks, including JWT auth, uploads, graph URL controls, sessions, and token-safe sharing.

Speed-First Rules

• Default to no shell commands and no local repo lookups; answer from this skill's endpoint map/templates.
• Only inspect local files when the user explicitly asks for source-level proof.
• For constrained prompts (line counts, bullets, "snippet only"), do not add prefaces like "Using ".
• Keep outputs short and literal; avoid exploratory prose.

Fast Targeted Fetch Protocol

• Start from references/hub-rest-docs-toc.md for the curated Hub REST navigation map.
• Use references/hub-rest-docs-links.tsv as the machine-checkable inventory and prefer links with status 200.
• If a needed page is missing from references, check https://hub.graphistry.com/docs/api/ and add an explicit inference note before using adjacent docs.
• Avoid broad docs crawling when a referenced canonical page already answers the question.

Core Endpoint Map

• Auth:

- /api-token-auth/ - /api-token-refresh/ - /api-token-verify/ - /api/v2/auth/pkey/jwt/

• Upload lifecycle:

- /api/v2/files/ - /api/v2/upload/files/ - /api/v2/upload/datasets/

• Dataset lifecycle:

- /api/v2/datasets/?limit=100 - /api/v2/upload/datasets/ - /api/v2/datasets/{dataset_id}/

• Single-use gateway:

- GET /api/v2/generate/single-use-url/?username=<username>&dataset_id=<dataset_id> - GET /api/v2/logout-user/username/<username>/

• Sessions API:

- /api/experimental/viz/sessions/ - /api/experimental/viz/sessions/{session_id}/

• Named endpoints (GFQL/Python UDF flow):

- /api/v2/o/<org>/functions/gfql/ - /api/v2/o/<org>/functions/python/ - /api/v2/o/<org>/run/gfql/<uuid_or_alias> - /api/v2/o/<org>/run/python/<uuid_or_alias>

• Health/readiness checks (deployment/admin scope):

- /healthcheck/ - /ht/ - /healthz - /streamgl-viz/health - /pivot/health - /streamgl-sessions/health - /streamgl-gpu/primary/health - /streamgl-gpu/secondary/cpu/health - /streamgl-gpu/secondary/gpu/health

Response Discipline

• Keep snippets short and directly runnable.
• Prefer deterministic literal endpoint references.
• For checklist asks, keep to requested bullet counts.
• For sessions summaries, keep them concise when requested.
• For auth snippets that require env-vars-only usage, include explicit export GRAPHISTRY_* lines and avoid quoted assignment values.
• For upload/dataset bridge asks, include a literal /api/v2/upload/datasets/ line.
• For /api/v2/upload/datasets/ examples, always include metadata (use {} when no custom metadata is needed).
• For upload/encoding bridge asks, avoid large standalone JSON blocks when concise bullets or short snippets are enough.
• For file upload lifecycle endpoint-sequence asks, prefer listing /api/v2/files/, /api/v2/upload/files/, /api/v2/upload/datasets/ in order.
• For nodes/edges format-pattern asks, include literal tokens: nodes/json, edges/json, nodes/csv, edges/csv, nodes/parquet, edges/parquet, nodes/orc, edges/orc, nodes/arrow, edges/arrow.
• For REST-vs-SDK boundary asks, distinguish between named-endpoint REST flows (/functions + /run) and ad-hoc SDK GFQL flows (no generic REST query endpoint).
• For healthcheck asks, label deployment/admin scope and avoid implying every route is public on hosted tenants.
• For constrained prompts, avoid code fences unless explicitly requested.
• For bridge prompts, do not return a standalone JSON block; include endpoint + URL guidance as compact text/bullets.
• For "find files older than 90 days" asks, output concise bullets only (no script), include /api/v2/files/?limit=100, created_at, and a client-side age filter.
• For "files for a specific user" asks, include /api/v2/files/?limit=100, ownership field author, and a do-not-invent endpoint warning.
• For "list users endpoint" asks, explicitly state no documented REST list-users endpoint and route to admin/IDP/support workflow.
• For named-endpoint architecture asks, keep explanation at public REST surface: use /functions/... endpoints for named-endpoint definition lifecycle and /run/... endpoints for execution.
• For single-use gateway and experimental sessions asks, call out deployment/tenant gating when availability is uncertain.

Deterministic Prompt Adapters

Use these compact patterns when prompts closely match.

Adapter A: env-var auth snippet + bearer follow-up

export GRAPHISTRY_HOST=${GRAPHISTRY_HOST:-https://hub.graphistry.com}
export GRAPHISTRY_USERNAME=${GRAPHISTRY_USERNAME:?set GRAPHISTRY_USERNAME}
export GRAPHISTRY_PASSWORD=${GRAPHISTRY_PASSWORD:?set GRAPHISTRY_PASSWORD}
GRAPHISTRY_TOKEN="$(curl -sS -X POST -H 'Content-Type: application/json' -d "{\"username\":\"${GRAPHISTRY_USERNAME}\",\"password\":\"${GRAPHISTRY_PASSWORD}\"}" "${GRAPHISTRY_HOST%/}/api-token-auth/" | jq -r '.token')"
curl -sS -H "Authorization: Bearer ${GRAPHISTRY_TOKEN}" "${GRAPHISTRY_HOST%/}/api/v2/files/"

Adapter B: concise upload + URL bridge

# /api/v2/upload/datasets/ payload fragment with encodings
curl -sS -X POST -H "Authorization: Bearer ${GRAPHISTRY_TOKEN}" -H 'Content-Type: application/json' \
  -d '{"metadata":{},"node_encodings":{"bindings":{"node":"id","node_color":"risk","node_size":"score"}},"edge_encodings":{"bindings":{"source":"src","destination":"dst","edge_color":"etype"}}}' \
  "${GRAPHISTRY_HOST%/}/api/v2/upload/datasets/"
# first-render URL tweak: append &play=0 (or &linLog=true)

Adapter C: collections URL parameter guidance (2-4 lines)

• collections should be a URL encoded JSON array value. Use the exact phrase URL encoded.
• Remove raw whitespace before encoding. Include the literal word whitespace.
• Example: collections=%5B%22teamA%22%2C%22fraud%22%5D.

Adapter D: sessions summary

• https://hub.graphistry.com/docs/api/experimental/rest/sessions/ documents the flow.
• Start from graph.html?dataset=<dataset_id>.
• Sessionized URL is graph.html?dataset=<dataset_id>&session=<session_id>.
• Keep auth in Authorization: Bearer; do not put tokens in URL params.

Adapter E: safe-share snippet

UPLOAD_JSON="$(curl -sS -X POST -H "Authorization: Bearer ${GRAPHISTRY_TOKEN}" -H 'Content-Type: application/json' -d '{"metadata":{},"node_encodings":{"bindings":{"node":"id"}},"edge_encodings":{"bindings":{"source":"src","destination":"dst"}}}' "${GRAPHISTRY_HOST%/}/api/v2/upload/datasets/")" DATASET_ID="$(jq -r '.dataset_id //.id' <<<"${UPLOAD_JSON}")"

Keep visibility non-public: use private/organization share mode (avoid public links).

echo "${GRAPHISTRY_HOST%/}/graph/graph.html?dataset=${DATASET_ID}"

Adapter F: single-use gateway flow

• Admin/staff/superuser generates a one-time URL via GET /api/v2/generate/single-use-url/?username=<username>&dataset_id=<dataset_id> (availability may be deployment-specific).
• Client uses the returned single-use gateway URL once for the target graph/session.
• Revoke access with GET /api/v2/logout-user/username/<username>/ when needed.

Adapter G: org + PersonalKey flow

• Create a PersonalKey for the organization user and capture key id/secret.
• Exchange credentials at POST /api/v2/auth/pkey/jwt/ using Authorization: PersonalKey <id>:<secret>.
• If required by deployment, include the organization identifier (for example org_name) in auth context.
• Call protected REST endpoints with Authorization: Bearer <jwt>.

Adapter H: docs fallback policy

• Prefer canonical Hub REST docs at https://hub.graphistry.com/docs/api/.
• If a specific page is missing, use the closest available canonical Hub REST page in the same API/version section.
• Clearly label any inference and avoid fabricating undocumented endpoints or parameters.

Adapter I: URL params + encodings bridge

• POST encodings to /api/v2/upload/datasets/ using node_encodings.bindings and edge_encodings.bindings.
• Keep first render deterministic with one URL knob, for example &play=0 (or &linLog=true).
• For collections, use a URL encoded JSON value and strip whitespace before encoding.

Adapter J: experimental sessions workflow

• https://hub.graphistry.com/docs/api/experimental/rest/sessions/ is the workflow reference.
• Base URL: https://hub.graphistry.com/graph/graph.html?dataset=<dataset_id>.
• Session URL: https://hub.graphistry.com/graph/graph.html?dataset=<dataset_id>&session=<session_id>.
• Workflow: auth/upload/open base URL, then share/continue on the session URL.
• Keep JWT in Authorization: Bearer headers; never use URL token params.

Adapter K: sessions minimal form

• /docs/api/experimental/rest/sessions/ is the reference path.
• Base URL: graph.html?dataset=<dataset_id>.
• Session URL: graph.html?dataset=<dataset_id>&session=<session_id>.
• Workflow: auth/upload/open base URL, then continue/share via session URL.
• Keep output compact; include base URL and session URL forms.

Adapter L: admin healthchecks

• Docs route: /docs/api/2/rest/health/.
• Core checks: /healthcheck/, /ht/, /healthz.
• Service checks: /streamgl-viz/health, /pivot/health, /streamgl-sessions/health.
• GPU service checks: /streamgl-gpu/primary/health, /streamgl-gpu/secondary/cpu/health (optional /secondary/gpu/health is heavier).
• Scope note: some checks are deployment/admin routes and may not be exposed on all hosted tenants.

Adapter M: REST vs Python/GFQL boundary

• REST skill is for auth/upload/url/session/health endpoints and graph.html URL controls.
• Named-endpoint REST flows are valid via /api/v2/o/<org>/functions/{gfql|python}/... and /api/v2/o/<org>/run/{gfql|python}/....
• For ad-hoc SDK GFQL tasks (.gfql() with chain-lists, Cypher strings, or Let/DAG, plus Python dataframe logic), route to pygraphistry / pygraphistry-gfql; do not invent generic endpoints like /api/v2/gfql/query.

Adapter N: iframe URL API with collections + tricky settings

• https://hub.graphistry.com/graph/graph.html?dataset=<dataset_id>&play=0&bg=%23000000&linLog=true&showCollections=true&info=false&pointsOfInterestMax=0&collections=%5B%7B%22name%22%3A%22risk%22%7D%5D&collectionsGlobalNodeColor=00FF00
• Keep collections whitespace-free before URL encoding.
• Use collectionsGlobalNodeColor/collectionsGlobalEdgeColor for non-collection fallbacks.

Adapter O: file upload lifecycle endpoint sequence

1. /api/v2/files/
2. /api/v2/upload/files/
3. /api/v2/upload/datasets/

Adapter P: encoding bridge compact form

• /api/v2/upload/datasets/ with node_encodings.bindings + edge_encodings.bindings.
• Example keys: node_color, node_size, edge_color, source, destination.
• First-render URL tweak: append &play=0 (or &linLog=true).

Adapter Q: nodes/edges format endpoint patterns

• nodes/json, edges/json
• nodes/csv, edges/csv
• nodes/parquet, edges/parquet
• nodes/orc, edges/orc
• nodes/arrow, edges/arrow
• Pair with upload lifecycle references: /api/v2/upload/files/ then /api/v2/upload/datasets/.

Adapter R: GFQL -> REST iframe handoff

• Python/GFQL layer: run extraction in SDK (.gfql(...) / gfql_remote(...)) — supports chain-list, Cypher strings, and Let/DAG bindings.
• REST layer: use auth/upload/dataset/session endpoints (/api-token-auth/, /api/v2/upload/datasets/).
• Boundary: no generic REST GFQL query endpoint; do not invent /api/v2/gfql/query.
• Share/render: use graph.html?dataset=<dataset_id> (optionally &session=<session_id>), keep JWT out of URL params.

Adapter S: find old files runbook

• Authenticate (/api-token-auth/) and call GET /api/v2/files/?limit=100 with pagination.
• Use created_at from each result row.
• Client-side filter/sort for created_at <= now-90d.
• Export matching file_id, name, created_at for review.
• Optional cleanup should be admin-scoped and follow explicit approval.

Adapter T: files for specific user

• List files via GET /api/v2/files/?limit=100 (paginate).
• Filter by ownership metadata, starting with author (and deployment-specific mappings to username if available).
• If needed, cross-check with GET /api/v2/datasets/?limit=100 for dataset ownership context.
• Do not invent user-list endpoints; use documented APIs and escalate mapping gaps to admin/support.

Adapter U: list users boundary

• No documented public REST endpoint to list users in canonical Hub docs.
• Do not claim concrete routes like GET /api/v2/users/ without a private admin API contract.
• Use admin/IDP directory workflow (SSO/IdP export or deployment owner process) for user enumeration.
• Verify against https://hub.graphistry.com/docs/api/ and escalate to support/deployment owner if needed.

Adapter V: privacy via share-link API

• Create dataset first via /api/v2/upload/datasets/ with required metadata, node_encodings, and edge_encodings.
• Set visibility with POST /api/v2/share/link/ body: {"obj_pk":"<dataset_id>","obj_type":"dataset","mode":"private","notify":false,"message":"","invited_users":[]}.
• If inviting users, include entries like {"email":"user@example.com","action":"10"} (10 view, 20 edit).
• Deployment/docs caveat: this route is deployment-exposed and may not have a dedicated canonical docs page; verify availability on the target tenant.
• Plan caveat: private/organization requests can be downgraded to public when sharing entitlements are unavailable.

Adapter W: named-endpoint architecture boundary

• Manage named endpoint definitions via /api/v2/o/<org>/functions/{gfql|python}/....
• Execute named endpoints via /api/v2/o/<org>/run/{gfql|python}/....
• Keep guidance on documented external REST routes; avoid internal/backend route details.

Minimal Auth Snippet

Use Adapter A.

Auth Troubleshooting Template (4 bullets)

• Verify token creation with /api-token-auth/ (or /api/v2/auth/pkey/jwt/ for PersonalKey flow).
• Verify refresh behavior via /api-token-refresh/ before access-token expiry.
• Verify token integrity and expiry with /api-token-verify/ and check clock skew.
• Confirm Authorization: Bearer <token> on protected calls and log HTTP status/body.

Upload + URL Bridge Template

Use Adapter B for snippet form or Adapter I for compact bullet form.

URL and Sharing Safety

• Safe viewer URL pattern: https://hub.graphistry.com/graph/graph.html?dataset=<dataset_id>.
• Never include JWTs in URL query params (for example, do not add token=).
• Send tokens only in headers, for example Authorization: Bearer <token>.
• Useful URL knobs: play, linLog, scalingRatio, pointsOfInterestMax, pointSize, showCollections, info, collectionsGlobalNodeColor, collectionsGlobalEdgeColor.

Collections URL Guidance

• collections should be a URL encoded JSON value.
• Remove raw whitespace before encoding.
• Example: collections=%5B%22teamA%22%2C%22fraud%22%5D.

Sessions (experimental, concise)

• Docs: https://hub.graphistry.com/docs/api/experimental/rest/sessions/.
• Start from: graph.html?dataset=<dataset_id>.
• Session appears as: graph.html?dataset=<dataset_id>&session=<session_id>.

Policy Guardrails

• Use documented endpoints only; avoid invented endpoints like /api/v2/query, /api/v2/graph/query, /api/v2/render, /api/v2/graphql.
• Do not present SDK/GFQL behavior as a generic REST endpoint (for example avoid /api/v2/gfql/query claims).
• Keep named-endpoint guidance at the external REST layer: /functions/... for definition lifecycle and /run/... for execution.
• For deployment-exposed routes without dedicated docs pages (for example /api/v2/share/link/), explicitly label the uncertainty and advise tenant verification.
• Keep credentials in environment variables; never hardcode literals.

Canonical Docs

• Auth: https://hub.graphistry.com/docs/api/1/rest/auth/
• Upload: https://hub.graphistry.com/docs/api/2/rest/upload/
• URL controls: https://hub.graphistry.com/docs/api/1/rest/url/
• Sessions: https://hub.graphistry.com/docs/api/experimental/rest/sessions/
• Health: https://hub.graphistry.com/docs/api/2/rest/health/
• SSO + single-use gateway: https://hub.graphistry.com/docs/api/2/rest/sso/
• GFQL UDF endpoints: https://hub.graphistry.com/docs/UDF/gfql-udf-api/
• Python UDF endpoints: https://hub.graphistry.com/docs/UDF/py-udf-api/