Enterprise Code Search and Bug Investigation with Multi-Agent AI Systems

wix · Tech · 2026 · source Agentic: yes · promotional bias: med · summary-only (no fetched source) · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Enterprise code search and bug investigation

Wix needed to help developers search and understand code across thousands of repositories and services, then investigate bugs using internal technical context across GitLab, databases, logs, documentation, and related systems.

Constraints:

• The YouTube fetch is retryable, so the extraction relies on the ZenML summary.
• The organization has thousands of repositories and services.
• Code search must handle dependencies and complex codebase navigation.
• Bug investigation requires multiple technical systems and deep research.
• Token limits and response quality constrain multi-agent research.

Architecture

Wix built two connected systems: OctoCode, an MCP-based developer code-search tool, and Bilbo, an enterprise multi-agent bug-investigation service. OctoCode helps search repositories, understand dependencies, and navigate codebases; Bilbo orchestrates agents across GitLab, databases, logs, documentation, and internal systems for deep technical investigations.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  octocode["OctoCode"] --> repository-corpus["Repository corpus"]
  repository-corpus --> bilbo["Bilbo"]
  bilbo --> sub-agent-workers["Sub-agent workers"]
  bilbo --> internal-technical-systems["Internal technical systems"]
  internal-technical-systems --> sub-agent-workers
  sub-agent-workers --> context-management["Context management"]

Components

Data flow

Developers use OctoCode for repository search and dependency understanding; Bilbo receives a bug or research task, delegates work to multiple agents, integrates with GitLab, databases, logs, docs, and internal systems, and manages context and token limits while returning an investigation.

Why this architecture

Wix split search and investigation because codebase navigation and bug research have related but different needs: one requires fast developer search, while the other requires multi-agent orchestration across many internal systems.

Models used: MCP · LLMs · OctoCode · Bilbo · Sub-agents

Scale

• OctoCode has 90,000 downloads and 5,000 weekly active users.
  “OctoCode, is an MCP-based tool with 90,000 downloads and 5,000 weekly active users”summary, stated
• The systems target thousands of repositories and services.
  “searching and understanding code across thousands of repositories and services in a large organization.”summary, stated
• The presentation covers roughly 11 months of development.
  “The presentation covers two main systems developed over approximately 11 months: OctoCode, a developer-facing search tool, and Bilbo”summary, stated

Alternatives considered (roads not taken)

Use a single generic code assistant for both search and bug investigation.

Rejected because: Wix built separate OctoCode and Bilbo systems for developer search and deep multi-agent investigation.

Investigate bugs using code search alone.

Rejected because: Bilbo also integrates with logs, databases, documentation, GitLab, and other systems.

Let one model consume all investigation context directly.

Rejected because: The summary highlights sub-agent architectures and context management to handle token limits.

Patterns

multi-agent-orchestration (orchestration)

Bilbo orchestrates multiple agents for bug investigation and deep research.

tool-use-orchestration (orchestration)

The systems connect agents to GitLab, databases, logs, documentation, and internal systems.

context-compaction (context)

The architecture uses prompt engineering and context management to manage token limits during enterprise-scale research.

Tradeoffs

Build two systems instead of one.

Chose OctoCode for search and Bilbo for deep investigation. over A single undifferentiated assistant..

Rationale: Search/navigation and bug investigation require different orchestration and integration depth.

Use sub-agents for investigations.

Chose Multiple specialized agents. over Single-pass investigation by one model..

Rationale: The summary ties sub-agents to complexity, token-limit management, and response quality.

Gotchas (what bit them)

Enterprise code search spans too many repositories for simple navigation.

Root cause: Large organizations have thousands of repositories and services with dependencies.

Resolution: Provide an MCP-based code-search tool focused on repository search and dependency understanding.

Bug investigation requires more than source code.

Root cause: The relevant evidence lives in GitLab, logs, databases, docs, and internal systems.

Resolution: Use a service that orchestrates agents across the whole technical stack.

Token limits threaten investigation quality.

Root cause: Enterprise-scale code search and investigation produce too much context for a naive prompt.

Resolution: Use sub-agent architectures, custom tools, and context management.

Evaluation

The summary reports adoption metrics for OctoCode and qualitative system goals for response quality, dependency understanding, and investigation depth.

Metrics: 90,000 downloads · 5,000 weekly active users · Response quality · Token-limit management

Outcomes

• Wix built OctoCode for MCP-based enterprise code search.
• Wix built Bilbo for multi-agent bug investigation across internal systems.
• OctoCode reached 90,000 downloads and 5,000 weekly active users.

The case covers thousands of repositories/services, 90,000 downloads, 5,000 weekly active users, and two systems developed over approximately 11 months.