Guides

Connect compiled repository views to coding agents, then choose the smallest retrieval and planning surface that answers the task.

• Serve context

  MCP Server

  Build a reusable manifest and expose the full tool set, with results gated by fresh, available repository views.

  Connect an MCP client →

• Search without an index

  grep → Jev

  Let a model plan ripgrep searches over your checkout and rank the matches with Jev, using your OpenRouter account.

  Set up grep and Jev →

• Compose retrieval

  Agent skills

  Add bounded lexical, semantic, graph-navigation, and reranking skills to an agent runtime.

  Browse the skill surface →

• Embed CodeNib

  Reproduced agents

  Run LocAgent, Agentless, CoSIL, OrcaLoca, or RepoNavigator's jump tool with CodeNib's index as their backend.

  Choose an agent →

• Execute safely

  Isolated agent execution

  Run issue reproduction, edits, and tests through a provider-neutral, resource-bounded sandbox instead of the CodeNib service host.

  Design an isolated worker →

• Reproduce a deployment

  Reference deployments

  Reproduce a tested local stack while keeping repository loading, retrieval, and generation measurements distinct.

  Choose a reference deployment →

Recommended path

Use the MCP Server for the standard agent-facing workflow. Reach for Agent Skills when you need to compose a custom runtime, use Isolated Agent Execution before running repository code, and start from a Reference Deployment when reproducing a complete local stack. To choose between these routes, start with context methods and support matrix; the RAG Ops And Planner reference defines retrieval policy and planning boundaries.