Result types

This page lists every shape the HTTP API returns, and the request bodies that have a name. Field names are exactly as they appear on the wire. The same models exist in Python as remember.models.<Name> (the most used are also exported from remember); the remember client validates every response against them.

The shapes group into five families:

Family	Types	Returned by
Envelope	`Envelope` and its parts	Every read route except SQL queries, readiness, documents and deployment info; three of the four assured operations.
Context bundle	`ContextBundle/v2`	`combined_context`.
Query result	`QueryResult/v1`	`POST /query/sql`, `POST /query/sql/explain`, `POST /query/saved/{namespace}/{name}/run`.
Write and readiness	`IngestedVersion`, `PipelineReadinessReport`, `DocumentPage`	`POST /ingest`, `POST /readiness`, `GET /documents`.
Deployment	`ToolDescriptor`, `DeploymentBuildInfo`	`GET /operations`, `GET /deployment`.

Conventions

Required and nullable. "Nullable" means the field is present and may be null. Fields with a default are always present in responses.
Arrays. Array fields are always present; an empty result is [].
Unknown fields. Every model forbids fields it does not declare, so the remember client rejects a response with an unexpected key.

Timestamps

Timestamps in results are ISO 8601 strings in UTC. Fields typed "UTC date-time" below are checked to have an offset of exactly zero.

Warning

Send timestamps to GET /lookup/relations (valid_at) and to the graph routes (valid_at, believed_at) in UTC, with Z or +00:00. A timestamp with no offset or another offset is refused with 422 and a validation list naming the field; it is not converted. The assured operations' time object converts any offset to UTC for you.

Envelope

The result of every read that is not a SQL query: the answer plus its own account of what kind of answer it is, when it was true, how fresh it is, whether it was capped, and, when the answer is "no", which kind of no.

Which fields a route fills depends on the route; the rest stay empty. No HTTP route in this release fills ranking, changes, aggregate or pages.

Field	Type	Contents
`grain`	`Grain`	What kind of truth the result is.
`temporal_scope`	`TemporalScope`	The time the answer applies to.
`entities`	array of `EntityCandidate`	Resolve candidates.
`facts`	array of `FactResult`	Relations and observations.
`evidence`	array of `EvidenceResult`	Claims.
`fact_evidence`	array of `FactEvidence`	Which claim supports or contradicts which fact.
`evidence_totals`	array of `EvidenceTotal`	Exact evidence counts per fact and stance.
`chunks`	array of `ChunkEvidenceResult`	Source passages.
`sources`	array of `SourceRecord`	Documents.
`transcript`	array of `TranscriptEntry`	Decision history.
`nodes`	array of `GraphNode`	Entities (or documents) a traversal reached.
`paths`	array of `GraphPath`	Whole paths.
`edges`	array of `GraphEdge`	Relations (or citations) traversed.
`ranking`	array of `RankedItem`	A fused or reranked order.
`changes`	array of `ChangeRecord`	A change feed.
`aggregate`	`AggregateReport`, nullable	An enumerated aggregate.
`pages`	array of `PageRef`	Compiled pages.
`freshness`	`Freshness`	How current the answer's sources are.
`truncation`	`Truncation`, nullable	Present when a cap applied or the route always discloses one.
`dropped_by_hydration`	integer	Candidates that were nominated but no longer held when re-read. Default `0`.
`excluded_unstamped`	integer ≥ 0	Items left out because they carry no usable time. Default `0`.
`negative`	`Negative`, nullable	Set when the answer is a typed "no".

Which fields each route fills:

Route or operation	Grain	Fields
`GET /resolve`, `resolve_entity`	`fact`	`entities`, `truncation`, `negative`
`GET /lookup/relations`, `GET /lookup/observations`	`fact`	`facts`, `truncation`, `dropped_by_hydration`, `negative`
`GET /hydrate/relation/{relation_id}`	`composite`	`facts`, `evidence`, `sources`, `negative`
`GET /transcript/relation/{relation_id}`	`composite`	`transcript`, `truncation`, `negative`
Search routes	`evidence`	`evidence` or `chunks`, `dropped_by_hydration`, `negative`
Adjacent chunks	`evidence`	`chunks`, `dropped_by_hydration`, `negative`
Graph routes	`fact`	`nodes`, `paths`, `edges`, `truncation`, `negative`
`claims_and_sources_context`	`evidence`	`evidence`, `chunks`, `truncation`, `dropped_by_hydration`, `negative`
`facts_context`	`fact`	`facts`, `evidence`, `fact_evidence`, `evidence_totals`, `nodes`, `truncation`, `dropped_by_hydration`, `negative`

Grain

Value	Meaning
`fact`	What the memory holds true (relations, observations, entities, the graph).
`evidence`	What sources said (claims, chunks). Not a statement of truth.
`compiled`	A compiled page. Not returned by any HTTP route in this release.
`composite`	A fact together with the evidence and sources behind it, or a decision history.

TemporalScope

One of five shapes, told apart by mode. Every shape carries evaluated_at (when the read ran), believed_at (the memory's knowledge it read from) and identity_regime.

`mode`	Extra fields	Used by
`current`	none	Most reads; `time: {"mode": "current"}`.
`at`	`at` (UTC date-time)	`time: {"mode": "at"}`; lookups with `valid_at`.
`overlap`	`from`, `to` (UTC date-times, `to` ≥ `from`)	`time: {"mode": "overlap"}`.
`history`	none	`time: {"mode": "history"}`.
`as_of`	`valid_at` (UTC date-time)	Graph routes; names both clocks.

Field	Type
`mode`	string, as above
`evaluated_at`	UTC date-time
`believed_at`	UTC date-time
`identity_regime`	`IdentityRegime`, default `current`

IdentityRegime

Value	Meaning
`current`	Entities are identified as they are today (today's aliases and merges), even for a past instant. Every HTTP route uses this.
`as_of`	Entities are identified as they stood at the queried instant.

EntityCandidate

Field	Type	Contents
`entity_id`	UUID
`canonical_name`	string
`tier`	string	`T0` exact alias, `T1` trigram, `T2` phonetic, `T3` embedding.
`context_hits`	integer	Current relations linking this candidate to the `context_entity_ids` you sent. Default `0`.

FactResult

One relation or observation.

Field	Type	Contents
`fact_id`	UUID	The relation or observation id.
`kind`	string	`relation` or `observation`.
`label`	string	Human-readable statement of the fact.
`evidence_count`	integer	Claims supporting it.
`validity`	`Validity`
`temporal_match`	`TemporalMatch`	Default `possible`.
`contradiction_group`	UUID, nullable	The contradiction group it belongs to.
`contradiction`	`Contradiction`, nullable	The other sides, when the group is live.
`support`	`FactSupport`	Default `current`.

Validity

Field	Type	Contents
`valid_from`	UTC date-time, nullable	When it became true; `null` when unknown.
`valid_until`	UTC date-time, nullable	When it stopped being true; `null` when open or unknown.
`valid_precision`	`ClaimValidPrecision`	Default `unknown`.
`ingested_at`	UTC date-time	When the memory learned it.
`invalidated_at`	UTC date-time, nullable	When the memory stopped believing it.

ClaimValidPrecision

How exact a validity window is: unknown, instant, day, month, quarter, year, open (known start, still ongoing).

TemporalMatch

Value	Meaning
`confirmed`	The fact's validity certainly matches the requested time.
`possible`	It may match; for example, a bound is unknown.

FactSupport

Value	Meaning
`current`	At least one source still asserts it.
`withdrawn`	Every source that asserted it has stopped. The fact is still returned, flagged.

Contradiction

A fact's live contradiction, returned with all its sides up to a cap.

Field	Type	Contents
`group_id`	UUID
`co_members`	array of `CoMember`	The other sides, up to 25 inline.
`returned`	integer ≥ 0	Sides returned.
`total`	integer ≥ 0	Sides that exist.
`continuation`	string, nullable	Set when more sides exist than were returned.

CoMember

Field	Type
`fact_id`	UUID
`label`	string
`evidence_count`	integer
`validity`	`Validity`

EvidenceResult

One claim with its provenance.

Field	Type	Contents
`claim_id`	UUID
`doc_id`	UUID	The document it came from.
`chunk_id`	UUID	The chunk it was extracted from.
`claim_text`	string	The claim, rewritten to stand alone.
`source_span`	string	The source text it came from.
`char_start`	integer	Start offset of the span in the converted Markdown of its representation.
`char_end`	integer	End offset.
`evidence_spans`	array of `EvidenceSpan`	Every range of that chunk's representation that supports the claim.
`is_attributed`	boolean	The source reports someone else's statement ("Dana said …").
`is_current_testimony`	boolean	The claim still comes from a current version of its source.
`asserted_at`	UTC date-time, nullable	When the source made the statement. Relative phrases in `claim_text` are relative to this.
`claim_valid_from`	UTC date-time, nullable	Start of when the claim says it was true, as the source put it. `null` when precision is `unknown`.
`claim_valid_until`	UTC date-time, nullable	End of that window; `null` when open or unknown.
`claim_valid_precision`	string	`instant`, `day`, `month`, `quarter`, `year`, `open` or `unknown`. Default `unknown`.
`claim_valid_kind`	string, nullable	What the window describes: `event_time` (when an event happened), `effective_period` or `proposition_validity` (when a state was true), `measurement_period` (the period a figure covers).
`document_title`	string, nullable
`source_kind`	string, nullable
`corroboration_count`	integer ≥ 1, nullable	Claims with identical text grouped under this one (`claims_and_sources_context`).
`grouped_claim_ids`	array of UUID	The ids of the grouped claims.

EvidenceSpan

Field	Type	Contents
`char_start`	integer ≥ 0	Start of a half-open range.
`char_end`	integer ≥ 0	End; always greater than `char_start`.

FactEvidence

Field	Type
`fact_kind`	`relation` \| `observation`
`fact_id`	UUID
`claim_id`	UUID
`stance`	`supports` \| `contradicts`

EvidenceTotal

Field	Type	Contents
`fact_kind`	`relation` \| `observation`
`fact_id`	UUID
`stance`	`supports` \| `contradicts`
`returned`	integer ≥ 0	Claims of this stance in the result.
`total`	integer ≥ 0	Claims of this stance that exist. Never less than `returned`.

ChunkEvidenceResult

One source passage.

Field	Type	Contents
`chunk_id`	UUID
`doc_id`	UUID
`version_id`	UUID
`representation_id`	UUID	The converted reading of the version the offsets refer to.
`chunk_text`	string
`context_prefix`	string, nullable	Generated orientation text for the chunk, when one exists. Not evidence.
`char_start`	integer	Offsets in the representation's Markdown.
`char_end`	integer
`section_role`	string, nullable	The role of the section it sits in (`body`, `abstract`, `references`, …).
`document_title`	string, nullable
`source_kind`	string
`source_modified_at`	UTC date-time, nullable
`published_at`	UTC date-time, nullable

SourceRecord

One document.

Field	Type	Contents
`doc_id`	UUID
`title`	string, nullable
`source_kind`	string
`markdown_uri`	string, nullable	Where the converted Markdown is stored.
`mention_count`	integer ≥ 0, nullable
`first_mentioned_at`	UTC date-time, nullable
`last_mentioned_at`	UTC date-time, nullable

TranscriptEntry

One recorded decision.

Field	Type	Contents
`subject_kind`	string	`relation`, `observation`, `entity` or `k_page`.
`outcome`	string	What was decided.
`method`	string	How it was decided.
`confidence`	number, nullable
`related_id`	UUID, nullable	The counterpart: the other fact in a supersession, the entity absorbed in a merge.
`decided_by`	string
`decided_at`	UTC date-time
`features`	object, nullable	The inputs the decision recorded.

GraphNode

Field	Type	Contents
`entity_id`	UUID	The entity, or on citation paths the document.
`name`	string	Canonical name, or document title.
`hops`	integer ≥ 0	Distance from the start.

GraphEdge

Field	Type	Contents
`relation_id`	UUID	The relation, or on citation paths the cross-reference.
`subject_id`	UUID
`object_id`	UUID
`predicate`	string	On citation paths, the kind of reference.
`fact`	string, nullable	The relation's label, or the citation's context.
`evidence_count`	integer	`0` on citation paths.
`valid_from`	UTC date-time, nullable
`valid_until`	UTC date-time, nullable
`valid_precision`	`ClaimValidPrecision`	Default `unknown`.
`ingested_at`	UTC date-time, nullable
`invalidated_at`	UTC date-time, nullable
`support`	`FactSupport`	Default `current`.

GraphPath

A path is returned whole or not at all.

Field	Type	Contents
`length`	integer ≥ 1	Number of edges.
`nodes`	array of `GraphNode`, at least 2	In order.
`edges`	array of `GraphEdge`, at least 1	In order.

RankedItem

Field	Type
`item_id`	UUID
`score`	number
`signals`	object of string to number

ChangeRecord

Field	Type	Contents
`kind`	string	`relation`, `observation`, `claim` or `page`.
`change`	string	`new`, `invalidated`, `capped` or `recompiled`.
`id`	UUID
`label`	string, nullable
`at`	UTC date-time

AggregateReport

Field	Type
`form`	string
`buckets`	array of `AggregateBucket`
`total`	integer ≥ 0
`possible_total`	integer ≥ 0, default `0`
`bounded_by`	string, nullable

AggregateBucket

Field	Type
`key`	string, nullable
`count`	integer ≥ 0
`possible_count`	integer ≥ 0, default `0`
`entity_id`	UUID, nullable

PageRef

Field	Type
`artifact_id`	UUID
`page_kind`	string
`git_path`	string, nullable
`page_summary`	string, nullable
`last_compiled_at`	UTC date-time, nullable
`status`	string
`stale`	boolean, default `false`
`open_review_flags`	integer ≥ 0, default `0`
`redaction_required`	boolean, default `false`

Freshness

Field	Type	Contents
`pg_live_ts`	UTC date-time	The database time the answer was read at.
`p1_written_inline`	boolean	The search index is written in the same step as the database. Default `true`.
`p1_believed_at_horizon`	UTC date-time, nullable	The oldest `believed_at` the search index can answer; `null` means unbounded.
`k`	`KFreshness`, nullable	Present only when the answer used a compiled page.

KFreshness

Field	Type
`compiled_at`	UTC date-time, nullable
`stale`	boolean, default `false`
`open_flags`	integer ≥ 0, default `0`

Truncation

A cap is never silent: when one applies, this block says so.

Field	Type	Contents
`truncated`	boolean	More existed than was returned.
`returned`	integer ≥ 0	Items returned.
`estimated_total`	integer ≥ 0	Items seen before the cap.
`total_is_exact`	boolean	`false` when the count itself hit a cap. Default `true`.
`continuation`	string, nullable	A cursor to pass back. Only `POST /graph/neighborhood` accepts one.
`reason`	string, nullable	Which cap applied, when known: `resolve_candidate_limit`, `lookup_k_limit`, `result_budget`, `expansion_budget`, `frontier_budget`, `time_budget`, `depth_budget`.

Negative

A typed "no". Each kind asks for a different reaction.

Field	Type
`kind`	`NegativeKind`
`explanation`	string, at least 1 character
`workaround`	string, nullable

NegativeKind

Value	Meaning	What to do
`unknown_entity`	The name or id is not known (or no longer exists).	Resolve the name again, check spelling, or search claims and chunks.
`known_empty`	The thing exists, and nothing matches.	Treat it as a real "none"; broaden the query if you expected results.
`boundary`	The memory could not answer within a limit: an index not published, a graph not ready, a time budget spent.	Do not treat it as "none". Retry, or use a direct route.

Forgotten content is indistinguishable from content that never existed: it shows up as unknown_entity or known_empty.

ContextBundle/v2

The result of combined_context: two complete envelopes side by side.

Field	Type	Contents
`contract`	string	Always `ContextBundle/v2`.
`claims_and_sources`	`Envelope`	Always grain `evidence`.
`facts`	`Envelope`	Always grain `fact`.

QueryResult/v1

The result of every SQL statement route, whatever the outcome. It describes where the rows came from before giving them.

Field	Type	Contents
`contract`	string	Always `QueryResult/v1`.
`grade`	string	Always `exploratory_tabular`: no grain, negatives, contradiction completeness, exact totals or order are guaranteed beyond what the fields below state.
`request_id`	UUID
`deployment_id`	UUID
`surface_manifest_hash`	string	The query space version the statement ran against.
`query_space_schema`	string	Always `memory_v1`.
`query_hash`	string	Hash of the normalized statement and the types (never the values) of its parameters. Empty when the statement did not parse.
`query_language`	string	Always `sql`.
`saved_query`	object of string, nullable	For a saved query: `query_id`, `namespace`, `name`, `version`, `query_hash`.
`referenced_views`	array of string	Views the statement reads.
`referenced_functions`	array of string	Functions it calls.
`source_grain_tags`	array of string	The grain tags of the referenced views.
`columns`	array of `ResultColumn`
`rows`	array of arrays	One array per row, values in column order.
`returned_row_count`	integer ≥ 0
`returned_byte_count`	integer ≥ 0	JSON-encoded size of the returned rows.
`limits`	`ResultLimits`	The caps this statement ran under.
`truncated`	boolean
`truncation_reason`	string, nullable	`row_cap`, `byte_cap`, or a graph budget (`depth_budget`, `expansion_budget`, `frontier_budget`, `result_budget`, `time_budget`).
`exact_total_known`	boolean	Always `false` in this release.
`exact_total`	integer, nullable	Always `null` in this release.
`ordered_result`	boolean	The statement has a top-level `ORDER BY`.
`empty_result`	boolean	No rows were returned (also `true` on every rejection or failure).
`negative_kind`	null	Always `null`: SQL results never carry a typed negative.
`execution_started_at`	date-time
`evaluated_at`	date-time, nullable	Set only when every referenced view and function answers at one instant (`facts_current`, `graph_edges_current`, `contradiction_members_current`, `facts_as_of`).
`pg_snapshot_at`	date-time, nullable	The database snapshot time.
`elapsed_ms`	number ≥ 0
`termination_reason`	string	`completed`, `rejected` or `failed`.
`error_code`	string, nullable	One of the SQL query codes.
`error_message`	string, nullable	Safe to show; never raw database detail.
`warnings`	array of string	For example one entry per graph function that hit a budget.
`semantic_invocations`	array of `SemanticInvocation`	One per search function call.
`graph_invocations`	array of `GraphInvocation`	One per graph function call.

ResultColumn

Field	Type	Contents
`name`	string
`type`	string	The PostgreSQL type name (`text`, `uuid`, `bigint`, `timestamp with time zone`, …).
`nullable`	boolean	Always `true`: PostgreSQL does not report nullability for computed columns.

ResultLimits

Field	Type
`row_cap`	integer
`byte_cap`	integer
`statement_timeout_ms`	integer
`analytical_tier`	boolean (always `false` over HTTP)

SemanticInvocation

What one search function (semantic_*, lexical_*, fetch_chunk_bodies) did.

Field	Type	Contents
`function`	string
`nominated`	integer	Candidates the index returned.
`confirmed`	integer	Candidates that held when re-read.
`dropped_stale`	integer
`dropped_filtered`	integer, default `0`
`dropped_ambiguous`	integer, default `0`
`dropped_absent`	integer, default `0`
`dropped_body_mismatch`	integer, default `0`
`dropped_absent_current`	integer, default `0`
`dropped_absent_projection`	integer, default `0`
`dropped_hash_mismatch`	integer, default `0`
`policy_generation`	string, nullable
`embedder_generation`	string, nullable
`generation`	string, nullable
`pg_confirmed_at`	date-time, nullable
`termination_reason`	string, nullable

GraphInvocation

What one graph function did.

Field	Type
`ordinal`	integer ≥ 0
`function`	`graph_neighborhood` \| `graph_path` \| `graph_citation_path`
`truncated`	boolean
`truncation_reason`	`depth_budget` \| `expansion_budget` \| `frontier_budget` \| `result_budget` \| `time_budget`, nullable
`examined_edges`	integer ≥ 0
`returned_paths`	integer ≥ 0
`effective_depth`	integer ≥ 1
`effective_expansion_budget`	integer ≥ 1
`effective_frontier_budget`	integer ≥ 1
`effective_result_budget`	integer ≥ 1
`effective_time_budget_ms`	integer ≥ 1
`applied_valid_at`	date-time, nullable
`applied_believed_at`	date-time, nullable
`evaluated_at`	date-time, nullable

IngestedVersion

Returned by POST /ingest.

Field	Type	Contents
`deployment_id`	UUID
`doc_id`	UUID	The document.
`version_id`	UUID	The version; pass it to `POST /readiness`.
`content_hash`	string	SHA-256 of the bytes, hex.
`created`	boolean	`true` for a new version; `false` when the bytes matched the latest version and nothing new was stored.
`mime`	string, nullable	The MIME type recorded for these bytes, which conversion uses. Set by the first upload of the bytes, unless that type had no converter route and a later upload sent one that has.
`title`	string, nullable	The document's title, set by its first ingest.
`versioning_mode`	`snapshot` \| `living`, nullable	The document's versioning mode, set by its first ingest.
`parked`	`"no_route"` or `null`	`"no_route"` when the upload's conversion is parked waiting for a conversion route for its MIME type: the bytes are stored, and conversion waits until an operator adds a route if needed and runs `remember ops resume-no-route`, or the same bytes are sent again with a MIME type that has a route. `null` means only that it is not parked for `no_route`; read processing state from `POST /readiness`.

This release's engine always sets mime, title and versioning_mode; they are null only from an engine release that predates them.

Readiness

ReadinessRequirements

The require object of POST /readiness. All four fields are required.

Field	Type
`pipeline`	boolean
`p1`	boolean
`live_graph`	boolean
`p3`	boolean

PipelineReadinessReport

Field	Type	Contents
`ready`	boolean	Every required capability is ready.
`versions`	array of `VersionPipelineReadiness`	One per requested version.
`capabilities`	object: `pipeline`, `p1`, `live_graph`, `p3` → `CapabilityReadiness`	All four, required or not.
`document_binding_generation`	string, nullable	As in `DeploymentBuildInfo`.
`model_bindings`	object of string to string	As in `DeploymentBuildInfo`.
`build_revision`	string	As in `DeploymentBuildInfo`.

VersionPipelineReadiness

Field	Type	Contents
`version_id`	UUID
`ready`	boolean	Every stage is `succeeded` or `skipped` and finished.
`stages`	array of `PipelineStageReadiness`	One per expected stage.

PipelineStageReadiness

Field	Type	Contents
`stage`	string	The pipeline stage.
`component_version`	string	The version of that stage's component the deployment expects.
`status`	string	`missing`, `pending`, `running`, `succeeded`, `failed`, `dead_letter`, `skipped`.
`finished_at`	date-time, nullable
`defer_reason`	string, nullable	Why a `pending` or `failed` stage waits: `no_route` (no converter for the file's type), `budget` (a spend budget is used up), `scheduled` (due later), `retry_backoff` (a `failed` stage waiting for its next attempt). `null` for any other status, and for pending work that is simply queued.

missing means no work for that stage and component version exists yet. failed may be retried by the pipeline; dead_letter will not be.

CapabilityReadiness

Field	Type	Contents
`required`	boolean	You asked for it.
`ready`	boolean
`checked_at`	date-time
`reason`	string	`ready`, or why not (see Ingest).
`version`	string, nullable	`p3` only: the snapshot version.
`built_at`	date-time, nullable	`p3` only.
`published_at`	date-time, nullable	`p3` only.

Documents

DocumentPage

Field	Type	Contents
`documents`	array of `DocumentSummary`
`cursor`	string, nullable	Pass to the next request; `null` on the last page.

DocumentSummary

Field	Type	Contents
`doc_id`	UUID
`title`	string, nullable
`source_kind`	string	`upload` for a file sent without a source identity.
`source_uri`	string, nullable	The `source_ref`, for documents sent with one.
`first_seen_at`	date-time	Never changes.
`latest`	`DocumentVersionSummary`	The newest version that has not been deleted.
`serving`	boolean	Some version of this document is `ready` and not deleted.

DocumentVersionSummary

Field	Type	Contents
`version_id`	UUID
`version_no`	integer
`status`	string	`ingesting`, `converting`, `structuring`, `ready`, `failed`, `deleted`.
`ingested_at`	date-time
`error`	string, nullable	Why it failed.

A version status of ready means conversion and structuring are done. It does not mean claims and facts have been extracted; check readiness for that.

DocumentDeletion

What DELETE /documents/{doc_id} changed. The counts describe this call only.

Field	Type	Contents
`doc_id`	UUID	The deleted document.
`deleted_at`	date-time	When the document was deleted. For a call that finished a deletion another path started, when the document was first hidden.
`claims_retired`	integer	Claims that stopped counting as evidence.
`relations_closed`	integer	Relations closed because this document was their only support.
`observations_closed`	integer	Observations closed because this document was their only support.

ToolDescriptor

One assured operation, as GET /operations returns it.

Field	Type	Contents
`name`	string
`description`	string
`input_schema`	object	JSON Schema of the arguments. Closed (`additionalProperties: false`).
`result_schema`	object	JSON Schema of the result.
`result_contract`	string	`envelope` or `context_bundle_v2`.
`output_grain`	string, nullable	`fact`, `evidence`, or `null` for `combined_context`.
`answer_intent`	string	`identity`, `claims_and_sources`, `facts` or `combined_context`.
`mutates`	boolean, nullable	Whether the operation changes memory. `false` on all four shipped operations.
`version`	integer ≥ 1, nullable
`implementation_plan_hash`	string of 64 hex characters, nullable	Identifies the exact plan.

DeploymentBuildInfo

Field	Type	Contents
`build_revision`	string	Source revision the image was built from; empty when unknown. Default `""`.
`model_bindings`	object of string to string	Model per pipeline role. Default `{}`.
`document_binding_generation`	string, nullable
`tools`	object of string to integer	MCP memory tool name to tool version, for every tool the deployment serves. Default `{}`.

Request bodies

SearchRequest

Body of POST /search/claims and POST /search/chunks.

Field	Type	Required	Default	Constraints
`query`	string	yes		1 to 4,096 characters.
`k`	integer	no	`10`	1 to 400.
`channel`	`semantic` \| `bm25`	no	`semantic`

AdjacentChunksRequest

Body of POST /chunks/adjacent.

Field	Type	Required	Default	Constraints
`chunk_id`	UUID	yes
`window`	integer	no	`1`	1 or 2.

The graph, readiness, SQL and saved-query bodies are described with their routes: Graph, Ingest, SQL queries.