Research Agent

Role

The Research Agent is responsible for executing structured research missions in the background. It uses a file-based task queue to receive work, runs a multi-phase research pipeline (the Mission DAG), stores typed intermediate artifacts at each phase, and produces a final Markdown report.

Status: Implemented (task queue, mission runner, phase DAG, and artifact pipeline operational)

Architecture note

Some capabilities of the Research Agent were initially prototyped in a shared runtime with the Media Agent. The architecture models them as separate logical agents because their responsibilities, tools, and risk profiles are different. The Research Agent is focused on structured information gathering and synthesis; the Media Agent is focused on entertainment automation and media control.

Why the Research Agent is separate

The Research Agent operates independently from the Orchestration Agent for two reasons:

Resource separation. Research missions are resource-intensive and potentially long-running. Keeping the Research Agent separate ensures that a running research mission does not compete with or block active development work.

Scope separation. Research mission execution is a well-defined, repeatable workflow with its own tools, artifacts, and state. Keeping it separate enforces clean boundaries and makes each component easier to reason about and debug.

Responsibilities

Mission flow (summary)

1. A task definition is written to the task queue directory as a JSON file.
2. The Mission Runner detects the task and creates an isolated run directory.
3. Phases execute in the order defined by the Mission DAG, respecting dependencies.
4. Each phase reads from prior artifacts and writes its output to the run directory.
5. The final phase aggregates all available artifacts into final_report.md.
6. The state file is updated to reflect completion or any errors.

Relationship with the Orchestration Agent

The Orchestration Agent can initiate research missions by writing task files to the Research Agent's task queue. Research reports are then available for the Orchestration Agent or the user to consume. The agents do not share a session, memory, or compute context.

A current limitation

The current research pipeline handles structured, repeatable research domains well (discovery, reviews, pricing, transcripts). Support for broader research domains — academic literature, technical documentation, multi-hop reasoning — is planned but not yet implemented.

Relationship to OpenClaw

The Research Agent is an OpenClaw-based implementation of a structured background research mission. OpenClaw provides the mission boundary, runtime model, state tracking, and artifact conventions; the Research Agent defines the concrete research pipeline, phase DAG, search adapters, and report-generation behavior.

For the framework-level execution model, see OpenClaw Framework Overview.

Related pages

→ Agentic Architecture — wider architecture context for research, tooling, memory, and audits → OpenClaw Framework Overview — the shared runtime model
→ Research Framework — pipeline components and mission flow details → Research Mission DAG — phase dependency graph → Research Artifacts — mission output files and lifecycle → Media Agent — separate agent for entertainment automation → System Overview — Research Agent in the full architecture