Base URL is whatever you pass to serve --addr (default http://127.0.0.1:8080). All request/response bodies are JSON. Write endpoints require Authorization: Bearer <token> and are fully audited.

{
  "id": "S-1-5-21-1004336348-1177238915-682003330-1103",
  "kind": "User",
  "name": "JDOE",
  "domain": "omenhound.lab",
  "sam_account_name": "jdoe",
  "distinguished_name": "CN=JDOE,OU=Users,DC=omenhound,DC=lab",
  "props": { "enabled": "true", "admincount": "1" }
}

kind ∈ User | Group | Computer | Domain | OU | GPO | Container | Trust | Base.

{ "source": "S-1-5-...-1103", "target": "S-1-5-...-512", "kind": "GenericAll",
  "props": { "inherited": "true" } }

{ "nodes": [ /* Node */ ], "edges": [ /* Edge */ ],
  "meta": { "collector": "ldap", "version": "…", "domain": "…",
            "started_at": "…", "finished_at": "…", "warnings": ["…"] } }

{ "status": "ok", "version": "f7e8055", "credential_ops_enabled": false,
  "write_api_enabled": true, "time": "2026-10-10T12:18:03Z" }

{ "nodes": 1284, "edges": 3610 }

Returns the entire graph as a Graph object (see above). This is what the UI loads.

Paginated node list. limit default 1000, max 10000; offset default 0.

{ "total": 1284, "limit": 1000, "offset": 0, "nodes": [ /* Node */ ] }

Same shape for edges:

{ "total": 3610, "limit": 1000, "offset": 0, "edges": [ /* Edge */ ] }

Returns a single Node, or 404 {"error":"node not found"}.

{ "outgoing": [ /* Edge */ ], "incoming": [ /* Edge */ ] }

Renders the dynamic abuse instruction for a concrete edge. The read field is the verify/enumerate step, command the act step (labeled Read / Add in the UI); both are interpolated from the two nodes' real properties, and missing values become <unknown> plus a warnings[] entry.

{
  "edge": { "source": "S-1-5-...-1103", "target": "S-1-5-...-512", "kind": "GenericAll" },
  "source": { /* Node */ },
  "target": { /* Node */ },
  "summary": "Full control over the target object …",
  "read": "impacket-dacledit -action read -target-dn CN=jdoe,OU=Users,DC=omenhound,DC=lab 'omenhound.lab/jdoe:<password>'",
  "command": "bloodyAD --host dc01.omenhound.lab -d omenhound.lab -u jdoe -p '<password>' set password 'Domain Admins' 'NewPassw0rd!42'",
  "alternates": ["impacket-dacledit …"],
  "opsec": "unicodePwd over LDAPS (4724) …",
  "event_ids": [4724, 5136],
  "warnings": ["target has no samaccountname"]
}

Errors: 400 if any of src/kind/dst is missing, 404 if a node or template is not found.

{ "edges": ["GenericAll", "WriteDacl", "MemberOf", "AdminTo", "…"] }

The grouped, edge-agnostic technique library — 10 categories / 54 techniques (33 edge-backed + 21 standalone). Commands keep their {PLACEHOLDER} tokens and are reference material only; read-only. See Technique Library for the full walkthrough.

{ "categories": [ {
  "id": "acl", "title": "Access control (ACL)", "note": "These are graph edges…",
  "techniques": [ {
    "id": "GenericAll", "title": "Full control over the object", "edge": "GenericAll",
    "summary": "Full control over the target object …",
    "read": "impacket-dacledit -action read -target-dn {TARGET_DN} '{DOMAIN}/{SOURCE_SAM}:<password>'",
    "command": "bloodyAD --host {DC} -d {DOMAIN} -u {SOURCE_SAM} -p '<password>' set password {TARGET_SAM} 'NewPassw0rd!42'",
    "alternates": ["impacket-dacledit …"],
    "opsec": "unicodePwd over LDAPS (4724) …",
    "event_ids": [4724, 5136],
    "requires": "Type 1 (relationship): requires {SOURCE_SAM} credentials.",
    "tags": ["acl", "write"]
  } ]
} ] }

Shortest routes between a source and a target. POST takes the request as JSON; GET takes the same fields as query parameters (from, to, direction, max_depth, max_paths). Read-only — the graph is never modified.

{
  "source_id": "S-1-5-21-…-1103",
  "target_id": "S-1-5-21-…-512",
  "direction": "outgoing",
  "max_depth": 6,
  "max_paths": 10,
  "include_node_types": ["User", "Group"],
  "exclude_node_types": [],
  "include_relationship_types": ["MemberOf", "GenericAll"],
  "exclude_relationship_types": [],
  "include_domains": ["corp.local"],
  "exclude_domains": [],
  "alternative": false,
  "required_intermediate_node_ids": []
}

{
  "paths": [
    {
      "nodes": [{"id": "S-1-…", "kind": "User", "name": "jdoe"}, {}],
      "relationships": [{"source": "S-1-…", "target": "S-2-…", "kind": "MemberOf"}],
      "length": 1
    }
  ],
  "truncated": false
}

The search returns up to max_paths distinct shortest routes of equal length — not every possible route. With alternative: true the same walk continues past the minimum hop count to enumerate longer simple routes, still in increasing length order; the caller can label the first (shortest) routes apart from the rest. Relationships are always returned in their stored orientation (source → target), even for incoming/undirected traversal, so a caller can match them against the dataset. truncated is true when a search limit (max paths, or the backend's expansion budget) stopped the walk early.

Lists the canned attack-path queries.

{
  "queries": [
    {
      "id": "kerberoastable",
      "title": "Kerberoastable users",
      "category": "Credentials",
      "description": "Enabled users with a service principal name (SPN)…",
      "cypher": "MATCH (u:User) WHERE NOT u.name STARTS WITH 'KRBTGT' AND (u.hasspn = true OR u.serviceprincipalname IS NOT NULL) RETURN u"
    }
  ]
}

The cypher field is the equivalent BloodHound CE query, so the same question can be run against a CE-populated Neo4j graph.

Runs one canned query against the current graph. {id} is any id from the list above — for example kerberoastable, asreproastable, dcsync, laps-readable, unconstrained-delegation, constrained-delegation, rbcd, adminto, has-session, high-value-targets, admincount, sensitive-users, owned, password-never-expires.

{
  "query": { "id": "dcsync", "title": "DCSync-capable principals", "category": "Credentials" },
  "count": 1,
  "findings": [
    {
      "node": { "id": "S-1-5-...-1103", "kind": "User", "name": "REPLADMIN" },
      "reason": "Holds directory replication rights; can replicate credential hashes from the domain.",
      "evidence": ["GetChangesAll → UXLAB"],
      "related": ["S-1-5-21-9"]
    }
  ],
  "notes": []
}

notes is where honesty lives: if the graph contains no node carrying the attribute (or no relationship of the kind) that a query depends on, the note says so, so a count of 0 is never presented as a clean bill of health. Errors: 404 for an unknown query id.

All write endpoints require Authorization: Bearer <token>. With no --api-token configured they return 403 write API is disabled. A wrong token returns 401 unauthorized and is audited as a denied auth event.

Body: a Graph object. Size-bounded to 512 MiB, unknown JSON fields rejected, and the graph is validated (vocabulary, referential integrity, size) before it touches the store. Content-Type: application/json.

Query parameters:

Without domain the historical behaviour is kept: every domain present in the file is imported as-is. A scoped import never silently defaults to "all domains".

# merge (default) — import everything in the file
curl -sS -X POST http://127.0.0.1:8080/api/v1/ingest \
  -H "Authorization: Bearer $OMENHOUND_API_TOKEN" \
  --data-binary @graph.json

# import only the corp.local objects, rejecting anything else
curl -sS -X POST "http://127.0.0.1:8080/api/v1/ingest?domain=corp.local&on_mismatch=reject&source=corp.json" \
  -H "Authorization: Bearer $OMENHOUND_API_TOKEN" \
  --data-binary @graph.json

# replace everything with this graph
curl -sS -X POST "http://127.0.0.1:8080/api/v1/ingest?mode=replace" \
  -H "Authorization: Bearer $OMENHOUND_API_TOKEN" \
  --data-binary @graph.json

{
  "nodes": 13, "edges": 14,
  "stats": { "nodes": 13, "edges": 14 },
  "tracked": true,
  "batch": { "id": "batch_20261013T134707.747_c4e579dc78", "domain": "corp.local", "status": "completed" }
}

tracked is true when the store records ingestion provenance (the in-memory store does). mode=replace with a domain clears only that domain first, so other domains survive.

Errors: 400 invalid/oversized/invalid graph, unknown mode, or a domain mismatch under on_mismatch=reject; 401 unauthorized; 403 disabled.

Dry run: validate an upload against a scope with the same domain, on_mismatch and unknown parameters and return what would be written. Nothing is stored.

{
  "preview": {
    "domain": "corp.local",
    "domains": ["corp.local"],
    "nodes": 13, "edges": 14,
    "assigned_nodes": 13, "stamped_nodes": 0,
    "mismatch_nodes": 0, "skipped_nodes": 0, "skipped_edges": 0,
    "cross_domain_edges": 0
  }
}

A domain mismatch under on_mismatch=reject returns 422 with the same preview plus an error string, so the UI can show exactly which objects would be refused.

Ingestion history, newest first. Each batch carries its id, source, domain, domains, mode, status (completed / partial / failed), created/updated counters, skipped counts and any errors. A batch id is provenance metadata, not an AD identity.

One batch's metadata. 404 if unknown.

Compute the impact of deleting everything a batch introduced, without deleting anything. The response is a DeletionPreview:

{
  "scope": "batch", "target": "batch_…",
  "delete_nodes": 13, "delete_edges": 14,
  "preserved_nodes": 0, "preserved_edges": 0,
  "shared_nodes": 0, "legacy_nodes": 0,
  "warnings": [],
  "token": "sha256:…"
}

Records that another import also needs (shared_nodes) are preserved; relationships incident to a deleted node are always removed (the model forbids dangling edges) and any forced removal is reported in warnings. token fingerprints the exact plan.

Delete a batch's data. Requires ?confirm=true and the token from a fresh preview; if the graph changed since the preview the request is rejected with 409 and a new preview is required. Returns a DeletionResult (deleted_nodes, deleted_edges, preserved_nodes, preserved_edges, batch_removed).

Ingested domains with their node/edge/batch counts.

Same preview-then-confirm workflow scoped to a whole domain. DELETE requires ?confirm=true plus the preview token. Only that domain's records are removed; records shared with another domain's import are preserved, and dependent relationships are reported before anything is committed.

Preview and delete an explicit selection within a scope (a batch or a domain):

{ "scope": "domain", "target": "corp.local",
  "node_ids": ["S-1-5-21-…-1103"], "edge_keys": ["S-1-…\u0000MemberOf\u0000S-2-…"] }

The backend validates every id against the scope and refuses ids that are not part of it, so a crafted request cannot delete outside the labelled scope.

Scoped deletion needs provenance. These endpoints work on any store that implements the IngestManager contract (the default in-memory store does). A store that cannot track provenance returns 501 with a clear message rather than performing an unscoped delete. The Neo4j backend currently does not implement provenance, so scoped deletion is unavailable there; use it with the in-memory store, or clear the whole database with DELETE /api/v1/graph on a database dedicated to OmenHound.

Delete every node and relationship the server currently holds. The write token is required, and a bare DELETE is refused: re-send with ?confirm=true (accepted values true, yes, 1) so a stray request cannot wipe the graph by accident.

curl -sS -X DELETE "http://127.0.0.1:8080/api/v1/graph?confirm=true" \
  -H "Authorization: Bearer $OMENHOUND_API_TOKEN"

{ "cleared": true, "stats": { "nodes": 0, "edges": 0 } }

Errors: 400 missing confirmation, 401 unauthorized, 403 disabled.

With --neo4j-uri the store treats the configured database as the graph, so this clears the whole database (all nodes). Point --neo4j-db at a database dedicated to OmenHound before using either the delete endpoint or mode=replace.

Stage a bind credential for a later credential-op. Requires --enable-credential-ops.

{ "domain": "corp.local", "username": "svc", "password": "…" }

201 Created:

{ "session_id": "…", "expires": "2026-10-10T12:33:03Z" }

The credential is memory-only, TTL-scoped, and never returned again.

{ "deleted": "…" }

Requires --enable-credential-ops and the header X-Omenhound-Approval: yes (otherwise 428 Precondition Required, audited as denied). The staged credential is taken (one-shot) and destroyed.

{ "target_dn": "CN=Jane,OU=Users,DC=corp,DC=local",
  "target_sid": "S-1-5-21-…-1103",
  "new_password": "…",
  "reason": "TICKET-123" }

200 OK:

{ "changed": "CN=Jane,OU=Users,DC=corp,DC=local", "target_sid": "S-1-5-21-…-1103" }

Errors: 403 credential ops disabled, 404 unknown session id, 502 AD write failed, 428 missing approval header.

All errors are {"error": "<message>"} with a matching HTTP status.

A per-IP token bucket applies to all requests; writes have a separate bucket (--write-rate-limit, --write-burst). Exceeding it returns 429. The client IP is taken from the socket and ignores X-Forwarded-For.

Next: Neo4j Schema & Attack Paths →