Guides Learn API Reference HTTP API

Functions

agent dynamicAgent / dynamic_agent tool extension mcp secret usd openThread / open_thread sqlite localMemory / local_memory localKnowledge / local_knowledge scriptedModel / scripted_model fakeSandbox / fake_sandbox host openTeam / open_team runEvals / run_evals assertNever / assert_never

Providers

otel postgres supermemory zep anthropic openai exa tavily brave e2b daytona modal mem0 slack github whatsapp

Agents and runs

Agent RunResult RunStream RunContext Tool McpServer Extension Hooks Skill Secret Store ConfigError ConfigErrorCode

Threads and evals

Thread CaseExpectation SavedCase

Host

Model adapters

Model ModelContext ModelInfo ModelRequest ModelResponse ModelChunk LookupResult LookupCapability

Sandbox adapters

Sandbox SandboxAuthority SandboxContext SandboxInfo SandboxSession ExecOutput ExecResult

Memory and knowledge adapters

MemoryProvider KnowledgeProvider Scope Binding MemoryRecord RecordRef MemoryHit KnowledgeSource DocVersion KnowledgeHit Doc SearchBackend SearchHit

Channel adapters

ChannelAdapter ChannelCapabilities RawRequest RawResponse VerifiedDelivery Inbound DeliveryOutcome

Other types

Exporter SyncReport SkippedBranch TeamAgent DynamicAgent TeamRunResult TeamRunStream Live EvalReport EvalCaseResult Verdict ModelBlockedError Principal TeamRef MemberRef Team StartRefusal InvalidDefinition SendRefusal AskRefusal ReplyRefusal ObserveRefusal CancelRefusal OperatorRefusal MemberState MemberResult StoredMemberResult StartResult SendResult AskOutcome AskResult ReplyResult Waited WaitResult MonitorResult CancelResult TeamStartResult TeamSendResult TeamAskResult TeamWaitResult TeamCancelResult TeamMember AskStatus TeamCursor TeamSource TeamItem StoreCorruptError

Live

What a live eval costs money with: the judge and a required budget.

Live evals need agents too.

Fields

judgeModelrequired

The model that grades each case against its rubric (judge.v1 instructions, one call per case).

budgetBudgetrequired

The run budget of every agent run and every judge run, subagents included. A run it stops is an error with budget_exhausted.

rubricreadonly string[]default []

Criteria added after each case's own.

TeamRunStream

What TeamAgent.stream() returns: a RunStream whose result is a TeamRunResult.

EvalReport

What runEvals returns and `threads eval --out` writes: per-case results, counts, the one-line summary, model calls and cost.

On this page