Skip to main content
Managed Research lets you hand a repo and research goal to hosted AI workers. Synth runs the workspace, tracks the run, and returns durable evidence: messages, tasks, logs, checkpoints, artifacts, usage, PRs, and final reports. Use it when the work should be repeatable and inspectable instead of trapped in a one-off chat.

​
Responsibility split

You provideSynth handles
Repo, goal, constraints, context, credentials, eval command, and review criteriaWorker orchestration, hosted workspace execution, durable state, logs, checkpoints, artifacts, usage, PRs, and final reports

​
Choose an interface

Get started (MCP + Codex / Claude)

Install the Managed Research MCP, set SYNTH_API_KEY, and verify with smr_list_projects.

MCP Quickstart

Start and steer runs from Codex, Claude Code, Cursor, or another MCP client.

Python SDK quickstart

Script runs, CI, and notebooks with SynthClient().research.

Run Configuration

Learn the launch fields for runbooks, work modes, harnesses, models, providers, and budgets.

Runs and Evidence

Inspect messages, task state, actor state, artifacts, checkpoints, usage, and reports.

​
Public contract

Managed Research and Research Factory are SynthClient().research products. Agent clients can use the hosted Managed Research MCP (smr_* tools). The web app at /smr is for interactive review. Install synth-ai[research] — not a separate managed-research package. Direct /smr REST usage is internal and unstable; use MCP or client.research for public integrations.

​
Run lifecycle

Most workflows follow the same loop:
  1. Create or select a project.
  2. Attach repo, context, credentials, data, and constraints.
  3. Preflight launch configuration before spending runtime.
  4. Start a one-off or project-scoped run.
  5. Steer the run with messages when needed.
  6. Inspect state, logs, tasks, artifacts, checkpoints, usage, and final report.