The flagship product of Autonomi -- an autonomous, spec-driven build system with a built-in trust layer. It does not call work done until it is verified: the RARV-C closure loop, 8 quality gates, the completion council, and the verified-completion evidence gate all have to clear before completion is accepted. Provider-agnostic: Claude Code, OpenAI Codex CLI, Cline, and Aider.
Spec-driven: turn a spec -- a PRD, GitHub issue, YAML feature file, or any natural-language brief -- into a fully deployed, production-ready application, with verification built in so "done" means verified, not just attempted.
Loki Mode is an enterprise-grade autonomous AI development orchestrator that:
- Built-in trust layer (verified completion) - Does not call work done until it is verified: the RARV-C closure loop, 8 quality gates, the completion council, and the verified-completion evidence gate all have to clear before completion is accepted
- Spec-driven development - Any spec (PRD, GitHub issue, OpenAPI/YAML doc, or one-line brief) drives the build end to end
- Executes complete SDLC phases - From requirements to deployment
- Manages multiple AI agents - Parallel execution with up to 10+ concurrent agents
- Provider-agnostic - runs on Claude Code (Tier 1), Cline (Tier 2), Codex / Aider (Tier 3 degraded); no vendor lock-in. Gemini CLI deprecated v7.5.18.
- MCP server - 34 tools (33 always available;
loki_memory_redactis gated onLOKI_MANAGED_AGENTS/LOKI_MANAGED_MEMORY) plus 3 resources and 2 prompts for integration with MCP-aware clients. Launch withloki mcp. - Learns across projects - Cross-project memory improves over time
- Provides enterprise controls - Authentication, audit logging, sandboxing
| Section | Description |
|---|---|
| [[Getting Started]] | Install and run your first session |
| [[CLI Reference]] | Complete command documentation |
| [[API Reference]] | REST API and WebSocket endpoints |
| [[Dashboard]] | Dark-themed web dashboard with council tab |
| [[Configuration]] | Config files and options |
| [[Environment Variables]] | All environment variables |
| [[Security]] | Security hardening and best practices |
| [[Enterprise Features]] | Auth, audit, registry, sandbox |
- Zero Configuration - Works out of the box with sensible defaults
- Spec to Production - Provide a spec (PRD markdown, GitHub issue, YAML brief), Loki handles the rest
- Provider-Agnostic - runs on Claude, Cline, Codex, or Aider, no vendor lock-in (Gemini deprecated v7.5.18)
- LSP Grounding - First-class agent tools for symbol verification via
mcp/lsp_proxy.py(v7.7.0+; pyright, typescript-language-server, gopls, rust-analyzer, jdtls; lsp_get_diagnostics regression fully fixed v7.7.14) - Cross-Project Learning - AI improves from every session
- Dark Dashboard - Vercel/Linear-inspired dark theme with sidebar navigation (replaces the deprecated VS Code extension as of v7.2.0)
- Token Authentication - Secure API access with scoped tokens
- Audit Logging - Compliance-ready JSONL audit trails
- Docker Sandbox - Isolated secure execution environment
- Project Registry - Multi-project orchestration
- Staged Autonomy - Approval gates for sensitive operations
- 8-Gate Quality System ([[Quality Gates]]) - Static analysis, test suite (pass/fail), blind 3-reviewer code review with severity blocking, anti-sycophancy Devil's Advocate, mock-integrity detection, test-mutation detection, documentation coverage (v6.75.0), and Magic Modules debate; plus a conditional backward-compatibility auditor (healing mode, v6.67.0, not numbered), the verified-completion evidence gate with inconclusive disclosure (v7.28.0), and held-out spec evals for anti-reward-hacking (v7.28.0)
- Guided First Build -
loki quickstart: four questions to a running build, with the real cost estimate shown before any spend, and a consent-gated Claude Code install offer when no provider is found (v7.29.0) - Completion Council - 3-member voting system with anti-sycophancy checks
- Security Hardening - Path traversal, XSS, injection, and memory leak protections
- TLS/HTTPS Dashboard - Encrypted API and dashboard connections
- OIDC/SSO Authentication - Enterprise identity provider integration
- RBAC Roles - Admin, operator, viewer, auditor role model
- Prometheus Metrics - OpenMetrics /metrics endpoint for monitoring
- Branch Protection - Agent sessions auto-create feature branches with PRs
- Log Integrity - SHA-256 chain-hashed tamper-evident audit entries (cross-file chain verification fixed v7.7.15: new
verify_all_logs()walks rotated daily logs in mtime order; previously verify_log_integrity false-negatived on any file beyond the first-ever) - Context Window Tracking - Real-time gauge, timeline, and per-agent breakdown of context usage
- Notification Triggers - Configurable alerts for context thresholds, task failures, budget limits
+------------------+ +------------------+ +------------------+
| Spec (PRD/Issue/ | --> | Loki Mode | --> | Deployed App |
| YAML brief) | | | | |
+------------------+ +------------------+ +------------------+
|
+-----------+-------------+-----------+
| | | |
+---------+ +---------+ +---------+ +---------+
| Claude | | Cline | | Codex | | Aider |
|(Tier 1) | |(Tier 2) | |(Tier 3) | |(Tier 3) |
+---------+ +---------+ +---------+ +---------+
|
+----+----+----+----+
| | | | |
Agent Agent Agent Agent (Parallel Execution)
| Channel | Command |
|---|---|
| npm | npm install -g loki-mode |
| Homebrew | brew install asklokesh/tap/loki-mode |
| Docker | docker pull asklokesh/loki-mode |
Current Version: 7.129.2 (CHANGELOG)
See [[Changelog]] for detailed release notes.
- GitHub Issues: Report bugs or request features
- Discussions: Community Q&A
This documentation is automatically updated with each release.