Greptile Feature Specifications

# ============================================================================
# GREPTILE - Complete Feature Specification
# ============================================================================
# Greptile is an AI-powered codebase intelligence platform that provides
# context-aware code review, semantic code search, and natural language
# querying over entire codebases using graph-based RAG.
#
# Company: Greptile Inc. (YC W24)
# Website: https://www.greptile.com
# Latest Version: v3 (Late 2025)
# ============================================================================

meta:
  name: Greptile
  tagline: "AI Code Review Agent with Complete Codebase Context"
  company: Greptile Inc.
  batch: Y Combinator W24
  founded: 2024
  series_a: "$25M (September 2025)"
  current_version: "v3"
  customers: "2,000+ organizations including NVIDIA, Brex, Coinbase, Stripe, Amazon"
  website: https://www.greptile.com
  documentation: https://docs.greptile.com
  api_base_url: https://api.greptile.com/v2/
  compliance: SOC 2 Type II

# ============================================================================
# CORE FEATURES
# ============================================================================

core_features:

  # --------------------------------------------------------------------------
  # 1. Codebase Indexing & Graph Construction
  # --------------------------------------------------------------------------
  codebase_indexing:
    description: >
      Greptile ingests entire repositories and builds a comprehensive
      knowledge graph of code entities and their relationships. This is
      the foundational layer that powers all other features.
    pipeline:
      - step: "AST Parsing"
        detail: >
          Parses every file's Abstract Syntax Tree to extract code entities
          including functions, classes, variables, files, and directories.
      - step: "Recursive Docstring Generation"
        detail: >
          Recursively generates docstrings for each AST node, producing
          semantic summaries that capture the intent and behavior of code.
      - step: "Embedding Generation"
        detail: >
          Embeds the generated docstrings into vector representations for
          semantic similarity search.
      - step: "Relationship Mapping"
        detail: >
          Connects all extracted entities by mapping function calls, imports,
          dependencies, variable usage, and inheritance hierarchies.
      - step: "Graph Persistence"
        detail: >
          Stores the complete graph for instant querying during code reviews,
          chat sessions, and API queries.
    entities_tracked:
      - functions
      - classes
      - variables
      - files
      - directories
      - imports
      - dependencies
    relationships_tracked:
      - function_calls
      - import_chains
      - dependency_trees
      - variable_usage
      - inheritance_hierarchies
      - module_relationships
    scaling:
      engine: Hatchet (workflow orchestration)
      capability: "Handles massive codebases (Linux kernel, CPython, VS Code)"
      indexing_time:
        small_repos: "3-5 minutes"
        large_repos: "1+ hours"
      reindexing: "Incremental; only re-indexes on changes"

  # --------------------------------------------------------------------------
  # 2. AI Code Review (PR Review Bot)
  # --------------------------------------------------------------------------
  code_review:
    description: >
      Greptile's flagship feature: an autonomous AI agent that reviews
      pull requests with full codebase context, not just the diff.
    version: v3
    agent_architecture:
      framework: "Anthropic Claude Agent SDK"
      behavior: "Autonomous multi-hop investigation"
      loop: >
        The agent runs in a loop with access to tools (codebase search,
        git history, learned rules). It has a high limit on inference/tool
        calls, enabling recursive search and multi-hop reasoning.
      tools_available:
        - codebase_search
        - git_history_access
        - learned_rules_lookup
        - dependency_tracing
        - pattern_comparison
    capabilities:
      - inline_comments: "Line-level comments on PR diffs identifying bugs, anti-patterns, and issues"
      - quick_fixes: "One-click fix suggestions for identified issues"
      - pr_summaries: "Natural language summaries of what changed and why"
      - sequence_diagrams: "Auto-generated call flow diagrams for PRs"
      - context_aware_suggestions: "Analyzes related files, APIs, configs, tests, docs, and history"
      - copyable_agent_prompts: "Each comment includes a prompt for coding agents with relevant context"
    investigation_actions:
      - "Examine git history to understand why code changed"
      - "Trace commits back to original pull requests"
      - "Compare code against patterns elsewhere in the repository"
      - "Follow nested function calls across files"
      - "Check dependencies and usage sites for impact analysis"
    detection_categories:
      - logic_errors
      - syntax_issues
      - style_violations
      - anti_patterns
      - repeated_code
      - security_vulnerabilities
      - performance_issues
      - missing_error_handling
    performance:
      catch_rate: "82% (41% higher than Cursor at 58%)"
      v3_vs_v2_acceptance: "70.5% higher acceptance rate"
      v3_context_usage: "~3x more context tokens than v2"
      v3_inference_cost: "75% lower for self-hosted (due to ~90% cache hit rate)"
    configuration:
      file: "greptile.json (placed in repo root)"
      parameters:
        labels: "Trigger reviews on specific PR labels"
        commentTypes:
          - logic
          - syntax
          - style
          - info
          - advice
          - checks
          - notes
        instructions: "Natural language custom guidance"
        special_instructions: "E.g., 'Include compliance implications section'"

  # --------------------------------------------------------------------------
  # 3. Custom Rules, Style Guides & Adaptive Learning
  # --------------------------------------------------------------------------
  custom_context:
    description: >
      A system for defining and enforcing team-specific coding standards,
      rules, and conventions that Greptile applies during reviews.
    rule_types:
      - name: "Custom Rules"
        description: "Natural language rules (e.g., 'use loops instead of list comprehension')"
        scoping:
          - by_repository
          - by_file_pattern  # e.g., "src/frontend/*.tsx"
      - name: "Style Guides"
        description: "Link existing style guide documents or markdown files"
        scoping:
          - by_repository
          - by_file_pattern
      - name: "Other Context"
        description: "Miscellaneous information Greptile may find useful during reviews"
        scoping:
          - by_repository
          - by_file_pattern
    adaptive_learning:
      description: >
        Greptile learns from team behavior to continuously improve review
        quality without manual rule creation.
      signals:
        - "Developer comments on PRs"
        - "Replies to Greptile comments"
        - "Thumbs up/down reactions on Greptile comments"
      management:
        dashboard: "Custom Context page in Greptile dashboard"
        auto_enable: "Toggle AUTO-ENABLE LEARNINGS switch"
    rule_analytics:
      - "Track whether rules are being used by Greptile"
      - "Track whether rules are being actioned by the team"

  # --------------------------------------------------------------------------
  # 4. Semantic Code Search
  # --------------------------------------------------------------------------
  semantic_search:
    description: >
      Natural language search across indexed repositories, powered by
      multi-modal retrieval combining semantic, keyword, and agentic search.
    search_modes:
      - name: "Semantic Similarity Search"
        description: "Vector-based search using embedded docstrings"
      - name: "Keyword Search"
        description: "Traditional text-matching search"
      - name: "Agentic Search"
        description: >
          An AI agent reviews the relevance of results and traces references
          to find important connections that pure search would miss.
    multi_repo: true
    result_types:
      - files
      - functions
      - classes
      - variables
      - code_snippets

  # --------------------------------------------------------------------------
  # 5. Natural Language Chat / Query
  # --------------------------------------------------------------------------
  chat_query:
    description: >
      Ask questions about your codebase in natural language and receive
      answers with references to specific files, functions, and code paths.
    interfaces:
      - web_app: "app.greptile.com"
      - vscode_extension: "Deprecated (maintainer wanted)"
      - cli: "npm i -g greptile"
    features:
      - multi_repo_context: "Query across multiple repositories in one session"
      - session_continuity: "Use session_id for follow-up questions"
      - streaming: "Real-time streaming responses"
      - genius_mode: "Enhanced query capabilities"

# ============================================================================
# API SPECIFICATION
# ============================================================================

api:
  base_url: "https://api.greptile.com/v2/"
  authentication:
    api_key:
      header: "Authorization: Bearer <GREPTILE_API_KEY>"
    github_token:
      header: "X-GitHub-Token: <GITHUB_PAT>"
      required_scope: "repo"

  endpoints:

    index_repository:
      method: POST
      path: /repositories
      description: "Submit a repository for indexing"
      request_body:
        remote:
          type: string
          enum: [github, gitlab]
          required: true
        repository:
          type: string
          format: "owner/repository"
          required: true
        branch:
          type: string
          required: true
        reload:
          type: boolean
          default: true
          description: "If false, won't reprocess if previously successful"
        notify:
          type: boolean
          default: true
          description: "Whether to notify upon completion"

    get_repository_status:
      method: GET
      path: "/repositories/{repositoryId}"
      description: "Check indexing status of a repository"
      parameters:
        repositoryId:
          format: "URL-encoded: remote:branch:owner/repository"
          example: "github%3Amain%3Aacme%2Fwidget"
      response:
        sha_present: "Indicates repository is indexed and ready to query"

    query_repository:
      method: POST
      path: /query
      description: "Query repositories in natural language"
      request_body:
        repositories:
          type: array
          items:
            remote: string
            repository: string
            branch: string
          required: true
        query:
          type: string
          required: true
        session_id:
          type: string
          description: "Continue a conversation"
        stream:
          type: boolean
          description: "Stream the response"
        genius:
          type: boolean
          description: "Enhanced query capabilities"

    search_repository:
      method: POST
      path: /search
      description: "Search repositories for code entities"

# ============================================================================
# INTEGRATIONS
# ============================================================================

integrations:

  source_control:
    - name: GitHub
      features: [pr_review, issue_enrichment, repository_indexing]
    - name: GitHub Enterprise
      features: [pr_review, issue_enrichment, repository_indexing]
    - name: GitLab
      features: [pr_review, repository_indexing]
    - name: GitLab Self-Hosted
      features: [pr_review, repository_indexing]
    - name: Bitbucket
      features: [repository_indexing]

  project_management:
    - name: Jira
      features: [issue_enrichment, context_source]
      protocol: MCP
    - name: Linear
      features: [issue_enrichment, context_source]

  documentation:
    - name: Notion
      features: [context_source]
      protocol: MCP
    - name: Google Drive
      features: [context_source]
      protocol: MCP

  communication:
    - name: Slack
      features: [codebase_qa, alert_routing]

  observability:
    - name: Sentry
      features: [alert_diagnosis, codebase_context_enrichment]
    - name: Datadog
      features: [alert_diagnosis, codebase_context_enrichment]

  automation:
    - name: Zapier
      features: [workflow_automation]

  ai_tooling:
    - name: MCP Server
      description: >
        Model Context Protocol server providing code search and querying
        to AI agents and IDE integrations.
      modes:
        - traditional_mcp: "Direct MCP client integration"
        - http_server: "JSON-RPC 2.0 interface for web/REST clients"
      tools_exposed:
        - index_repository
        - query_repository
        - search_repository
        - get_repository_info

# ============================================================================
# CLIENT INTERFACES
# ============================================================================

client_interfaces:

  web_app:
    url: "https://app.greptile.com"
    features:
      - "Chat with any indexed codebase"
      - "Dashboard for managing repos, rules, and analytics"
      - "Custom context management"
      - "Rule analytics and learnings"

  cli:
    install: "npm i -g greptile"
    features:
      - "Terminal-based codebase chat"
      - "Repository indexing"

  vscode_extension:
    status: "Deprecated (maintainer wanted)"
    marketplace: "Greptile.greptile-vscode"
    features:
      - "Natural language queries"
      - "Multi-repository search"
      - "Bug solving assistance"

  pr_bot:
    platforms: [GitHub, GitLab]
    trigger: "Automatic on PR creation or configurable via labels"

# ============================================================================
# LANGUAGE SUPPORT
# ============================================================================

language_support:
  fully_supported:
    - Python
    - JavaScript
    - TypeScript
    - Go
    - Elixir
    - Java
    - C
    - C++
    - "C#"
    - Swift
    - PHP
    - Rust
  additional:
    description: "Most other languages supported with slightly lower response quality"
    count: "30+ languages total"

# ============================================================================
# SECURITY & COMPLIANCE
# ============================================================================

security:
  compliance:
    - SOC_2_Type_II
    - "Annual external audits"
    - "Rigorous penetration testing"
  data_handling:
    code_storage: "Cloud product does NOT store customer code"
    retrieval: "Pulls snippets as needed from code host (GitHub/GitLab API)"
    self_hosted: "Full code stays within customer's infrastructure"
  authentication:
    - "GitHub OAuth login"
    - "Microsoft OAuth login"
    - "SSO/SAML (Enterprise)"
  enterprise_security:
    - "Self-hosted deployment in customer AWS"
    - "Bring-your-own-LLM"
    - "Complete data control"
    - "Custom DPA and terms of service"
    - "GitHub Enterprise support"

# ============================================================================
# PRICING
# ============================================================================

pricing:
  cloud:
    price: "$30/developer/month"
    billing: "Flat rate, unlimited reviews"
    trial: "14-day free trial"
    discounts:
      annual: "Up to 20% off for 1+ year contracts"
      startups: "50% off"
      open_source: "100% free"
  enterprise:
    price: "Custom annual contracts"
    deployment: "Self-hosted in customer AWS"
    features:
      - "Bring-your-own-LLM"
      - "SSO/SAML"
      - "Dedicated Slack support channel"
      - "Custom invoicing and payment terms"
      - "Custom DPA and terms of service"
      - "GitHub Enterprise support"

# ============================================================================
# VERSION HISTORY
# ============================================================================

version_history:
  v1:
    description: "Initial release - basic codebase indexing and search"
  v2:
    release: "2024"
    description: >
      Introduced code graph, multi-modal search, and PR review capabilities.
      Used a rigid flowchart-based orchestration for review steps.
    blog: "https://www.greptile.com/blog/greptile-2"
  v3:
    release: "Late 2025"
    description: >
      Complete rewrite of core agent architecture. Switched to Anthropic
      Claude Agent SDK for autonomous, multi-hop investigation. Catches
      3x more critical bugs than v2. 70.5% higher acceptance rate.
    key_changes:
      - "Anthropic Claude Agent SDK integration"
      - "Autonomous multi-hop investigation (no rigid flowchart)"
      - "Long-term memory for the review agent"
      - "MCP server for IDE and agent integration"
      - "Jira, Notion, Google Docs context via MCP"
      - "Highly scoped custom rules"
      - "Simplified pricing ($30/dev/month flat)"
      - "~90% prompt cache hit rate"
      - "75% lower inference costs for self-hosted"
    blog: "https://www.greptile.com/blog/greptile-v3-agentic-code-review"

# ============================================================================
# USE CASES
# ============================================================================

use_cases:
  primary:
    - name: "Pull Request Review"
      description: "Automated, context-aware code review on every PR"
    - name: "Codebase Q&A"
      description: "Ask natural language questions about any indexed repository"
    - name: "Bug Detection"
      description: "Identify bugs, anti-patterns, and security issues in PRs"
  secondary:
    - name: "Onboarding"
      description: "Help new team members understand codebase structure and conventions"
    - name: "Legacy Code Exploration"
      description: "Navigate and understand legacy codebases where original authors left"
    - name: "Documentation Generation"
      description: "Auto-generate wiki articles with codebase context"
    - name: "Alert Diagnosis"
      description: "Connect to Sentry/Datadog to diagnose production alerts with code context"
    - name: "Test Failure Diagnosis"
      description: "Integrate with CI to suggest fixes for failing tests"
    - name: "Ticket Enrichment"
      description: "Add codebase context to Jira/Linear issues automatically"
    - name: "Custom Internal Tools"
      description: "Build PR review bots, Slackbots, and documentation updaters via API"

# ============================================================================
# BENCHMARKS (July 2025 - Independent Evaluation)
# ============================================================================

benchmarks:
  source: "Independent evaluation, July 2025"
  catch_rate_ranking:
    - tool: Greptile
      catch_rate: "82%"
    - tool: Cursor
      catch_rate: "58%"
    - tool: GitHub Copilot
      catch_rate: "~55%"
    - tool: CodeRabbit
      catch_rate: "44%"
    - tool: Graphite
      catch_rate: "6%"
  tradeoff: >
    Greptile has the highest catch rate but also the highest false positive
    rate. Users get more real bugs detected alongside more noise.

# ============================================================================
# INDUSTRY METRICS (From Greptile's 2025 Data)
# ============================================================================

industry_metrics:
  period: "March - November 2025"
  median_pr_size_growth: "33% increase (57 to 76 lines changed per PR)"
  lines_per_developer_growth: "76% increase (4,450 to 7,839 lines/developer)"
  attribution: "AI coding tools acting as force multiplier"