Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.compose.market/llms.txt

Use this file to discover all available pages before exploring further.

Manowar lives under runtime/src/manowar. The modules are small enough to inspect directly, but this map gives you the ownership boundaries first.

Runtime Core

ModuleResponsibility
index.tsExposes the single framework id, manowar.
mode.tsResolves local/cloud host behavior and worker initialization gates.
runtime.tsRegisters agents and workflows, validates cards, derives wallets, warms runtimes.
framework.tsCreates agents, model wrappers, managed workers, streams, stop controllers, and Responses calls.
telemetry.tsExtracts tokens, records usage, feedback, learning records, and run metadata.

Agent Loop

ModuleResponsibility
agent/graph.tsNative model/tool loop with streaming, tool repair, and turn budgets.
agent/context.tsPer-turn execution context via AsyncLocalStorage.
agent/identity.tsAgent identity hydration and compact identity rendering.
agent/memory.tsPre-turn retrieval, post-turn persistence, and explicit remember calls.
agent/memory-scope.tsGlobal/local memory scope resolution.
agent/state.tsTool leases and visibility states: base, memory, tool, model, swarm.
agent/tools/*Tool factories, tool-call parsing, model/agent/search clients, and surface sentence.

Memory

ModuleResponsibility
memory/agent-loop.tspre_turn, post_turn, and remember workflow.
memory/layers.tsSix-layer retrieval across working, scene, graph, patterns, archives, vectors.
memory/vector.tsAtlas vector search, keyword fallback, access-count updates.
memory/graph.tsDurable fact extraction and graph-layer retrieval.
memory/embedding.tsVoyage embeddings through MongoDB’s AI gateway.
memory/ranking.tsTemporal decay, Cloudflare rerank, and MMR ordering.
memory/mongo.tsMongoDB connection, collections, and indexes.
memory/cache.tsRedis cache, embedding cache, and scoped invalidation.
memory/operations.tsConsolidation, patterns, archives, decay, cleanup, and IPFS sync.
memory/evals.tsMemory retrieval evaluation metrics.

Harness

ModuleResponsibility
harness/types.tsTyped MAL plan, step, budget, result, and subagent contracts.
harness/interpreter.tsPlan parser, validator, deterministic dispatcher, checkpoint/proof orchestration.
harness/engine.tsSubagent execution and budget enforcement.
harness/parallel.tsFanout concurrency and gather modes.
harness/registry.tsRegistered-agent validation through the API.
harness/scratchpad.tsPrivate per-agent/run scratchpad.
harness/conclave.tsShared swarm bus.
harness/checkpoint.tsRedis MAL checkpoints.
harness/proof.tsHash proof bundles and optional Pinata pinning.
harness/sandbox.tsDaytona-backed isolated subagent execution.
harness/coordinators.tsDynamic coordinator model discovery from the catalog.

Workflow

ModuleResponsibility
workflow/orchestrator.tsPlan -> Act -> Reflect workflow execution.
workflow/planner.tsStructured planning, review, reflection, validation, and graph ordering.
workflow/delegation.tsHTTP calls to embedded agent runtime.
workflow/context.tsContext window lookup and cleanup thresholds.
workflow/memory.tsWorkflow memory writes, search, reliability, and safe wipe summaries.
workflow/checkpoint.tsWorkflow observations, decisions, insights, errors, and persistence.
workflow/run-tracker.tsIn-memory run lifecycle and cron stats.
workflow/triggers.tsNatural-language trigger parsing and Temporal schedule management.
workflow/registry.tsWorkflow card loading and connector tool discovery.

Comparison Map

Manowar moduleClosest public-frame patternDifference that matters
Native executorOpenAI Agents SDK runner, ADK runnerManowar calls the Compose inference gateway and records metering evidence.
MAL harnessLangGraph graph, ADK workflow agentsMAL is a typed plan language interpreted by the runtime, not a graph library embedded in user code.
MemoryLangGraph store, ADK session/state/memoryManowar stores layered memory in first-party Mongo/Redis and injects compact context automatically.
Swarm delegationOpenAI handoffs, AutoGen messagesDelegation requires registered Compose agents with wallet identity.
Model storeOpenRouter-style catalog searchModel selection uses the Compose generated catalog and public model ids.
TelemetryOpenAI tracing, LangSmith, AutoGen logsTelemetry is in-runtime usage and feedback records used by Compose billing and learning surfaces.