Claude Platform Docs

Dream

Dream 생성
BetaDream Beta.Dreams.Create(parameters, cancellationToken = default)
POST/v1/dreams
Dream 목록 조회
DreamListPage Beta.Dreams.List(parameters, cancellationToken = default)
GET/v1/dreams
Dream 조회
BetaDream Beta.Dreams.Retrieve(parameters, cancellationToken = default)
GET/v1/dreams/{dream_id}
Dream 취소
BetaDream Beta.Dreams.Cancel(parameters, cancellationToken = default)
POST/v1/dreams/{dream_id}/cancel
Dream 보관
BetaDream Beta.Dreams.Archive(parameters, cancellationToken = default)
POST/v1/dreams/{dream_id}/archive
Models
class BetaDream { Type; ID; ArchivedAt; /* 11 more */ }

An asynchronous memory-consolidation job that reads a memory store plus a set of session transcripts and writes consolidated memories into an output memory store — a new store by default, or an existing store chosen via output_behavior. The Dreams API is in research preview: the request and response shapes are volatile and may change without the deprecation period that applies to generally-available endpoints.

class BetaDreamError { Type; Message; }

Failure detail for a Dream whose status is failed.

required string Type
required string Message
class BetaDreamInput: union
class BetaDreamMemoryStoreInput { Type; MemoryStoreID; }

An input memory store the dream reads from. The dream never mutates this store unless it is also the destination: with output_behavior {type: "update_existing"} the job consolidates this store in place.

required Type Type
required string MemoryStoreID
minLength1
class BetaDreamSessionsInput { Type; SessionIds; }

Input session transcripts the dream reads.

required Type Type
required IReadOnlyList<string> SessionIds
class BetaDreamMemoryStoreInput { Type; MemoryStoreID; }

An input memory store the dream reads from. The dream never mutates this store unless it is also the destination: with output_behavior {type: "update_existing"} the job consolidates this store in place.

required Type Type
required string MemoryStoreID
minLength1
class BetaDreamMemoryStoreOutput { Type; MemoryStoreID; }

An output memory store the dream writes consolidated memories into.

required Type Type
required string MemoryStoreID
class BetaDreamModelConfig { ID; Speed; }

Model identifier and configuration applied to every pipeline stage. Same wire shape as the Agents API ModelConfig.

required string ID

Model identifier, e.g. "claude-opus-5". 1-256 characters.

minLength1
maxLength256
Speed Speed

Inference speed mode. fast provides significantly faster output token generation at premium pricing. Not all models support fast; invalid combinations are rejected at create time.

One of the following:
Standard("standard")
Fast("fast")
class BetaDreamModelConfigParam { ID; Speed; }

Model identifier and configuration applied to every pipeline stage.

required string ID

Model identifier, e.g. "claude-opus-5". 1-256 characters.

minLength1
maxLength256
Speed? Speed

Inference speed mode. fast provides significantly faster output token generation at premium pricing. Not all models support fast; invalid combinations are rejected at create time.

One of the following:
Standard("standard")
Fast("fast")
class BetaDreamOutput { Type; MemoryStoreID; }

An output memory store the dream writes consolidated memories into.

required Type Type
required string MemoryStoreID
class BetaDreamSessionsInput { Type; SessionIds; }

Input session transcripts the dream reads.

required Type Type
required IReadOnlyList<string> SessionIds
enum BetaDreamStatus

Lifecycle status of a Dream.

Pending("pending")
Running("running")
Completed("completed")
Failed("failed")
Canceled("canceled")
class BetaDreamUsage { CacheCreationInputTokens; CacheReadInputTokens; InputTokens; OutputTokens; }

Cumulative token usage for the dream across every pipeline stage.

required int CacheCreationInputTokens

Total tokens used to create prompt-cache entries (sum of all TTL tiers).

formatint32
required int CacheReadInputTokens

Total tokens read from prompt cache.

formatint32
required int InputTokens

Total uncached input tokens consumed across every pipeline stage.

formatint32
required int OutputTokens

Total output tokens generated across every pipeline stage.

formatint32
class BetaDreamingError: union
class BetaOutputBehavior: union
class BetaOutputBehaviorCreateNew { Type; }

The default destination: the job creates a new output memory store as a clone of the memory_store input and writes the consolidated memories into it. The input store is never mutated.

required Type Type
class BetaOutputBehaviorUpdateExisting { Type; MemoryStoreID; }

The job writes the consolidated memories into this existing memory store instead of creating one. In EAP the store must be the job's own memory_store input, so the job consolidates the store in place.

required Type Type
required string MemoryStoreID
minLength1
class BetaOutputBehaviorCreateNew { Type; }

The default destination: the job creates a new output memory store as a clone of the memory_store input and writes the consolidated memories into it. The input store is never mutated.

required Type Type
class BetaOutputBehaviorUpdateExisting { Type; MemoryStoreID; }

The job writes the consolidated memories into this existing memory store instead of creating one. In EAP the store must be the job's own memory_store input, so the job consolidates the store in place.

required Type Type
required string MemoryStoreID
minLength1
class BetaTargetStoreHeldError { Type = "conflict_error"; Message; }

The output_behavior.memory_store_id target is still held by a prior {type: "update_existing"} dream — one that is pending or running, or was canceled with its final writes still landing. Rarely the named dream has just finished (completed/failed) and its execution is still closing; an immediate retry then almost always succeeds. The message names the holding dream when the server can identify it (rarely omitted); poll it to a terminal state or cancel it, then retry. Carried with x-should-retry: false.

JsonElement Type = "conflict_error"
string Message

Human-readable description of the conflict, naming the dream that holds the target store when the server can identify it.