# Autonomi - Full Context for LLMs

> Autonomi is a platform for autonomous AI agent systems. Our flagship product, Loki Mode, takes a PRD and autonomously builds, tests, and deploys complete software products with minimal human intervention.

## Company

- Name: Autonomi
- Website: https://www.autonomi.dev
- Contact: founder@autonomi.dev
- GitHub: https://github.com/asklokesh
- Discord: https://discord.gg/k8NpBhc5KA
- LinkedIn: https://www.linkedin.com/company/autonomi-dev-agents
- Founded: 2025

## Loki Mode - Autonomous Agent System

Loki Mode is a source-available (BUSL-1.1; free for personal, internal, and academic use), vendor-independent CLI tool. The released v7 engine draws on 41 specialized agent roles across 8 domains to autonomously build software from a PRD, OpenSpec change specification, or GitHub/GitLab/Jira issue. The CLI is free forever -- you pay only your AI provider API costs. Current stable version: v7.129.3.

The current next-engine branch previews 48 roles across 9 groups by adding 7 executive roles. The v8 runtime branch uses the raw Anthropic SDK for one-shot calls and Claude Agent SDK query() for the RARV loop. Both are unreleased branch previews. The SDK route is opt-in, fails closed, and preserves .loki state, dashboard, cost, council, gate, and evidence-receipt contracts. Raw SDK calls use HTTPS; the Agent SDK still uses its bundled Claude Code runtime binary.

### How It Works

1. User provides input: a PRD (loki start ./prd.md), OpenSpec spec (loki start --openspec PATH), or issue (loki start ISSUE)
2. Loki Mode classifies complexity and activates the RARV cycle (Reason, Act, Reflect, Verify)
3. The orchestrator adopts 41 specialized agent roles across 8 domains through up to 8 SDLC phases
4. Each iteration passes through 8 deterministic quality gates plus an advisory Secure gate, with a blind-review council and anti-sycophancy
5. A 3-agent completion council votes on project completion (2/3 majority required)
6. The finished product is deployed with minimal human intervention

### Three Input Pathways

1. PRD-based: loki start ./my-prd.md (freeform Markdown, greenfield projects)
2. Spec-driven: loki start --openspec PATH (OpenSpec Bridge, v6.11.0, brownfield with delta specs)
3. Issue-driven: loki start ISSUE (GitHub, GitLab, Jira, Azure DevOps)

### Agent Architecture

41 specialized agent roles across 8 domains:

1. Engineering Domain (8 roles): eng-frontend, eng-backend, eng-database, eng-mobile, eng-api, eng-qa, eng-perf, eng-infra
2. Operations Domain (8 roles): ops-devops, ops-sre, ops-security, ops-monitor, ops-incident, ops-release, ops-cost, ops-compliance
3. Business Domain (8 roles): biz-marketing, biz-sales, biz-finance, biz-legal, biz-support, biz-hr, biz-investor, biz-partnerships
4. Data Domain (3 roles): data-ml, data-eng, data-analytics
5. Product Domain (3 roles): prod-pm, prod-design, prod-techwriter
6. Growth Domain (4 roles): growth-hacker, growth-community, growth-success, growth-lifecycle
7. Review Domain (3 roles): review-code, review-business, review-security
8. Orchestration Domain (4 roles): system-level coordination

### SDLC Phases

8 phases in complex tier:
1. Bootstrap - Environment setup and dependency installation
2. Discovery - Requirements analysis and user story generation
3. Architecture - System design and technology selection
4. Infrastructure - CI/CD, containerization, and deployment setup
5. Development - Parallel code generation across all modules
6. QA - Automated testing, security scanning, and performance benchmarks
7. Deployment - Production deployment with health checks
8. Growth - Analytics, onboarding, and initial content

### Complexity Tiers

- Simple: 3 phases, for single-file scripts and utilities
- Standard: 5 phases, for typical web apps and APIs
- Complex: 8 phases, for enterprise systems with 10+ files

### Quality Gates

8 deterministic gates plus advisory Secure gate:
1. Static Analysis - CodeQL, ESLint, type checking (multi-language)
2. Test Suite - pass/fail, with mock-integrity and test-mutation checks
3. Blind Review - 3-reviewer system (code, business, security) with severity blocking
4. Anti-Sycophancy - Devil's advocate on unanimous reviews
5. Mock Integrity - catches excessive or hollow mocking
6. Test Mutation - validates test quality via mutation analysis
7. Documentation Coverage - validates docs alongside code
8. Magic Modules Debate - cross-persona debate on module changes
Advisory: Secure-by-default gate (vuln scan, secrets, supply chain) runs advisory by default and can be flipped to blocking via policy.

### Multi-Provider Support (Bring Your Own LLM)

Loki Mode supports 4 AI providers with no vendor lock-in:

- Claude: Tier 1 (full support, parallel agents, task tool, MCP)
- Cline: Tier 2 (subagents + MCP, near-full capabilities)
- Codex (OpenAI): Tier 2 (near-full capabilities)
- Aider: Tier 3 (18+ providers via litellm: OpenRouter, Ollama, Together AI, GROQ, DeepSeek, Azure, Bedrock, and more)

Note: Gemini was deprecated in v7.5.18 due to Google upstream changes.

Dynamic model resolution with 3-tier precedence:
1. Provider-specific env var (e.g., LOKI_CLAUDE_MODEL_DEVELOPMENT)
2. Generic tier env var (e.g., LOKI_MODEL_DEVELOPMENT)
3. Provider default

Per-phase model selection: Use expensive models for planning, cheaper models for development, fastest models for quick tasks.

Budget control: LOKI_BUDGET_LIMIT (USD spend cap), LOKI_MAX_TIER (caps model cost tier).

### Enterprise Features

- Hash-chained audit trails: SHA-256 chain in .loki/audit/audit.jsonl, tamper-evident, loki audit verify
- Compliance: SOC 2 Type II, ISO 27001, GDPR compliance reports
- Policy engine: 4 enforcement points (pre_execution, pre_deployment, resource, data), ALLOW/DENY/REQUIRE_APPROVAL
- Data scanning: Detects API keys, tokens, PII (SSN, credit cards, emails, phone numbers)
- RBAC: 4 roles (admin, operator, viewer, auditor), OIDC integration
- OpenTelemetry: OTLP/HTTP+JSON export, spans for Project/Task/Phase/Gate/Agent/Council
- MCP server/client for tool integration
- A2A event bus (Python/TypeScript/Bash) for inter-agent communication
- Jira/Linear/GitHub sync for issue tracking

### Open Source Pricing

- License: BUSL-1.1 (source-available; free for personal, internal, and academic use)
- Platform cost: the CLI is free forever (no subscription, no usage fees)
- You pay only your AI provider API costs
- All features included: dashboard, quality gates, audit trails, OTEL, policy engine
- No cloud dependency: runs entirely on your machine
- Budget control: LOKI_BUDGET_LIMIT caps provider spend
- Token economics tracking: loki memory economics

### Installation

Primary: npm install -g loki-mode
Alternative: git clone https://github.com/asklokesh/loki-mode.git ~/.claude/skills/loki-mode
Homebrew: brew install asklokesh/tap/loki-mode
Docker: docker pull asklokesh/loki-mode

### Usage

Three input pathways:
- loki start ./my-prd.md (PRD-based)
- loki start --openspec PATH (spec-driven, v6.11.0)
- loki start ISSUE (issue-driven)

Other commands:
- loki doctor (diagnostics)
- loki status (session status)
- loki dashboard (open web UI)
- loki agent list (41 agent roles)
- loki worktree list (progressive isolation)
- loki audit lint (compliance check)
- loki telemetry status (OTEL status)
- ~99 CLI commands total

### OpenSpec Bridge (v6.11.0)

OpenSpec (YC W26) is a spec-driven development framework. Loki Mode v6.11.0 bridges OpenSpec change directories into the execution pipeline:
- Adapter: openspec-adapter.py (827 lines, stdlib-only Python)
- Delta spec injection: ADDED/MODIFIED/REMOVED annotations into agent prompts
- GIVEN/WHEN/THEN scenario mapping to quality gate acceptance criteria
- Automatic complexity classification based on task count, spec files, and design.md
- 28 unit tests, 6 test fixtures

### BMAD Method Integration (v6.1.0)

BMAD Method handles phases 1-3 (Analysis, Planning, Solutioning). Loki Mode handles phase 4+ (Implementation). Combined workflow produces higher-quality PRDs (7.5-8.5/10 vs 5-6/10 for freeform PRDs).

## Website Pages

- Homepage (https://www.autonomi.dev/): Overview of Autonomi and Loki Mode
- Product (https://www.autonomi.dev/loki-mode): Detailed product page with features, architecture, and quick start
- Use Cases (https://www.autonomi.dev/use-cases): BYO LLM, enterprise governance, PRD-to-production, open source pricing
- Use Case - BYO LLM (https://www.autonomi.dev/use-cases/bring-your-own-llm): Vendor independence, 4 providers, 18+ models, dynamic resolution
- Use Case - Enterprise Governance (https://www.autonomi.dev/use-cases/enterprise-governance): Audit trails, compliance, RBAC, policy engine, OTEL
- Use Case - PRD to Production (https://www.autonomi.dev/use-cases/prd-to-production): End-to-end autonomous delivery, 41 agent roles, 8 phases, 8 gates plus advisory Secure
- Use Case - Open Source Pricing (https://www.autonomi.dev/use-cases/open-source-pricing): BUSL-1.1 source-available license, free-forever CLI, pricing comparison
- Documentation (https://www.autonomi.dev/docs): 25+ pages of technical documentation
- Blog (https://www.autonomi.dev/blog): Release notes and engineering updates
- About (https://www.autonomi.dev/about): Company mission, philosophy, and team
- Labs (https://www.autonomi.dev/labs): Experimental projects and research