VibeCody Competitive Roadmap v3

Date: March 2026 Previous: ROADMAP-v2.md (February 2026) — all phases complete Scope: Futureproofing gaps identified in FIT-GAP-ANALYSIS-v5.md; 12 new phases across 4 priority tiers

Current State

All phases from Roadmap v1 (1-5) and v2 (6-9) are complete. FIT-GAP v4’s 23 gaps are all closed (Phases 53-67). VibeCody has:

Metric Count
Unit tests ~5,745
Skill files 539+
AI providers 17 direct + OpenRouter (300+)
VibeUI panels 128+
REPL commands 65+
Gateway platforms 18
Rust modules 100+

Phase 10: MCP Ecosystem & Scalability

10.1 MCP Lazy Loading / Tool Search (P0)

Why: Claude Code v2.1.74 showed 95% context reduction through deferred tool loading. As MCP ecosystems grow (30+ plugins in Cursor alone), eager loading of all tool schemas at startup becomes unsustainable.

Deliverables:

  • mcp_lazy.rs — LazyToolRegistry with manifest-only startup loading
  • ToolSearch: keyword search across all MCP servers, loads matching tools on demand
  • LRU eviction: auto-unload unused tool schemas after configurable idle timeout
  • Config flag: mcp.lazy_loading = true|false (default: true)
  • Metrics: track context savings, cache hit rates, load times
  • Tests: 30+ unit tests

10.2 MCP Verified Plugin Directory (P1)

Why: Cursor has 30+ verified MCP partner plugins. VibeCody needs a curated directory to compete.

Deliverables:

  • mcp_directory.rs — PluginDirectory with categories, ratings, verification status
  • Plugin manifest format: mcp-plugin.toml
  • Verification pipeline: checksum, permission audit, sandboxed test
  • McpDirectoryPanel.tsx — browse, search, install, rate
  • REPL: /mcp install|search|update|uninstall|verify
  • Tests: 25+ unit tests

Phase 11: Context & Collaboration

11.1 Context Bundles / Spaces (P0)

Why: Copilot Spaces allow teams to share curated context sets. No open-source equivalent exists. This is a differentiation opportunity.

Deliverables:

  • context_bundles.rs — ContextBundle struct (pinned files, instructions, excludes, model prefs)
  • Bundle file format: .vibebundle.toml (portable, versionable)
  • Operations: create, activate, deactivate, share, import, export
  • Multiple active bundles with priority ordering
  • Auto-injection into agent system prompt
  • ContextBundlePanel.tsx — create/edit/share/browse team bundles
  • REPL: /bundle create|activate|list|share|import
  • Tests: 35+ unit tests

11.2 Usage Metering / Credit System (P1)

Why: Credit-based billing is replacing flat subscriptions (Cursor, Devin ACU). Even for BYOK users, teams need per-user/per-project cost allocation.

Deliverables:

  • usage_metering.rs — UsageMeter with per-agent/task/user token tracking
  • Credit budgets: configurable limits per team/user/project with alerts
  • Usage reports: daily/weekly/monthly by provider, model, task type
  • Chargeback: allocate AI costs to projects/departments
  • Integration with cost observatory
  • UsageMeteringPanel.tsx — dashboards, budgets, alerts, reports
  • Tests: 30+ unit tests

Phase 12: Cloud & Infrastructure

12.1 Cloud Provider Deep Integration (P0)

Why: Amazon Q Developer’s deep AWS integration is a moat. VibeCody needs equivalent coverage for AWS/GCP/Azure to serve enterprise teams.

Deliverables:

  • cloud_providers.rs — CloudProviderManager with AWS/GCP/Azure adapters
  • IAM policy generation: analyze code → detect service usage → generate least-privilege policies
  • IaC template generation: CloudFormation, Terraform, Pulumi from project structure
  • Service cost estimation from usage patterns
  • Cloud-specific skill files (20+): aws-lambda, gcp-cloud-run, azure-functions, etc.
  • REPL: /cloud aws|gcp|azure with subcommands
  • Tests: 40+ unit tests

12.2 Browser-Based Mode (P2)

Why: Bolt.new, Replit, and Devin all offer zero-install web experiences. VibeCody should offer a lightweight web mode for onboarding and demos.

Deliverables:

  • Compile vibe-core to WASM (text buffer, search, file system abstraction)
  • Static web app with Monaco editor, terminal emulator, file tree
  • WebContainer or server-side sandbox for command execution
  • Feature subset: chat, basic agent, file editing, preview
  • Progressive enhancement: web → desktop migration path
  • Hosted at vibecody.dev (or similar)

Phase 13: Protocol & Interoperability

13.1 Agent Client Protocol (ACP) Support (P1)

Why: Zed is pushing ACP as a standard for agent-editor communication. Early adoption ensures VibeCody can participate in multi-agent ecosystems.

Deliverables:

  • acp_protocol.rs — ACP message types, capability negotiation, tool registration
  • ACP server mode: expose VibeCody tools to ACP-compatible editors
  • ACP client mode: connect to external ACP agents as tool providers
  • Dual-protocol support: VibeCody native + ACP
  • Protocol version negotiation and graceful fallback
  • Tests: 25+ unit tests

13.2 JetBrains Agent Hooks (P2)

Why: Copilot’s JetBrains GA includes agent hooks. VibeCody’s JetBrains plugin should match.

Deliverables:

  • Extend JetBrains plugin with hook registration API
  • Events: pre-edit, post-edit, pre-commit, post-commit, agent-start, agent-complete
  • Hook types: shell command, HTTP webhook, LLM-based
  • Settings UI in JetBrains for hook configuration
  • Tests: 15+ unit tests

Phase 14: Quality & Enterprise Readiness

14.1 SWE-bench Benchmarking Harness (P2)

Why: SWE-bench scores are the primary marketing metric in AI coding. VibeCody should let users benchmark their own provider+config.

Deliverables:

  • swe_bench.rs — BenchmarkRunner with task download, agent execution, scoring
  • Support SWE-bench Verified, SWE-bench Pro, custom suites
  • Provider/model comparison across runs
  • BenchmarkPanel.tsx — run, compare, export reports
  • REPL: /benchmark run|compare|export
  • Tests: 20+ unit tests

14.2 Long-Session Memory Profiling (P2)

Why: Claude Code has fixed memory leaks in long sessions. VibeCody’s 8+ hour batch runs need similar safeguards.

Deliverables:

  • session_memory.rs — MemoryProfiler with periodic heap sampling
  • Leak detection: track allocation growth rate, flag anomalies
  • Auto-compact: evict stale context, compress history
  • Session health dashboard with memory usage graphs
  • Tests: 15+ unit tests

14.3 SOC 2 Technical Controls (P3)

Why: Enterprise procurement increasingly requires SOC 2 / ISO 27001. Technical controls can be built now even if certification is later.

Deliverables:

  • compliance_controls.rs — ControlInventory mapped to SOC 2 Trust Service Criteria
  • Immutable audit trail for all AI-generated code changes
  • Access control documentation: RBAC policies, key rotation, session management
  • Data retention policies: configurable log retention, PII redaction
  • Compliance report generation for auditors
  • Tests: 20+ unit tests

14.4 Multi-Modal Unified Agent (P3)

Why: Voice (voice.rs), vision (providers), and code (agent.rs) currently operate as separate features. Unifying them enables more natural developer interaction.

Deliverables:

  • multimodal_agent.rs — UnifiedAgent processing mixed inputs per turn
  • Voice → text → agent action pipeline
  • Screenshot → vision → code generation pipeline
  • Interleaved conversation context (voice + images + code)
  • Auto-detection of input mode
  • Tests: 25+ unit tests

Priority & Timeline Summary

Phase Feature Priority Estimated Effort
10.1 MCP Lazy Loading P0 2-3 days
10.2 MCP Plugin Directory P1 3-4 days
11.1 Context Bundles P0 2-3 days
11.2 Usage Metering P1 2-3 days
12.1 Cloud Provider Integration P0 4-5 days
12.2 Browser-Based Mode P2 2-3 weeks
13.1 ACP Protocol P1 2-3 days
13.2 JetBrains Agent Hooks P2 1-2 days
14.1 SWE-bench Harness P2 3-4 days
14.2 Memory Profiling P2 1-2 days
14.3 SOC 2 Controls P3 3-4 days
14.4 Multi-Modal Agent P3 4-5 days

P0 total: ~8-11 days (3 features) P1 total: ~7-10 days (3 features) P2 total: ~3-5 weeks (4 features) P3 total: ~7-9 days (2 features)

Success Criteria

After completing Phases 10-14, VibeCody will:

  1. Scale MCP — Handle 100+ MCP servers without context bloat (lazy loading)
  2. Match Copilot Spaces — Open-source context bundle standard (.vibebundle.toml)
  3. Compete with Amazon Q — Cloud provider integration for AWS/GCP/Azure
  4. Interoperate — ACP compliance for multi-agent ecosystem participation
  5. Enterprise-ready — Usage metering, SOC 2 controls, compliance reports
  6. Benchmarkable — Built-in SWE-bench runner for transparent evaluation
  7. Accessible — Browser-based mode for zero-install onboarding

Competitive Position After v3

With all v3 phases complete, VibeCody would be the only tool that combines:

  • Open-source + self-hostable + 17 AI providers
  • CLI + Desktop IDE + Browser (3 surfaces)
  • 539+ domain skills across 25+ industry verticals
  • MCP lazy loading + verified plugin directory
  • Context bundles (open standard)
  • Cloud provider deep integration (AWS/GCP/Azure)
  • ACP protocol compliance
  • SWE-bench benchmarking
  • SOC 2 technical controls
  • Multi-modal unified agent (voice + vision + code)
  • 18-platform messaging gateway
  • Batch generation (3M+ lines) + multi-QA validation
  • Air-gapped deployment with Ollama

No competitor currently offers more than 4-5 of these in combination.