Roboticus has no cryptocurrency token. Any token using this name is unaffiliated with this project.

Roboticus is an autonomous agent runtime built in Rust as a single optimized binary. It features streaming LLM responses, heuristic model routing, 3-level semantic caching with SQLite persistence, hybrid RAG retrieval (FTS5 keyword + vector cosine with optional HNSW ANN indexing), multi-provider embeddings (OpenAI/Ollama/Google with n-gram fallback), zero-trust agent-to-agent communication, 4-layer prompt injection defense, a dual-format skill system, human-in-the-loop approval workflow, headless browser tool integration, Context Observatory (efficiency metrics, outcome grading, cost attribution, optimization recommendations), response transform pipeline, addressability filtering for group chats, flexible network binding, and built-in financial management with DeFi yield optimization. All twelve crates compile into one process with no IPC overhead — inter-component communication is direct async function calls on the tokio runtime, backed by a single SQLite database (39 tables including FTS5).

brew tap robot-accomplice/roboticus
brew install roboticus

winget install -e --id RobotAccomplice.Roboticus

cargo install roboticus-server

macOS / Linux:

curl -fsSL https://roboticus.ai/install.sh | bash

Windows (PowerShell):

irm https://roboticus.ai/install.ps1 | iex

Then initialize and start:

roboticus init
roboticus serve

# Homebrew
brew upgrade roboticus

# Winget
winget upgrade RobotAccomplice.Roboticus

# Self-update (any install method)
roboticus update binary --method download --yes

The workspace is organized as twelve crates with a strict dependency hierarchy:

flowchart TD
    subgraph EntryPoints ["Entry Points"]
        SERVER["roboticus-server<br/>(HTTP, dashboard, WS)"]
        TESTS["roboticus-tests<br/>(integration tests)"]
    end

    subgraph Orchestration ["Orchestration"]
        SCHED["roboticus-schedule<br/>(heartbeat, cron)"]
    end

    subgraph DomainLogic ["Domain Logic"]
        AGENT["roboticus-agent<br/>(loop, tools, policy)"]
        WALLET["roboticus-wallet<br/>(ethereum, x402, yield)"]
    end

    subgraph Infrastructure ["Infrastructure"]
        DB["roboticus-db<br/>(SQLite, migrations)"]
        LLM["roboticus-llm<br/>(client, format, routing)"]
        CHANNELS["roboticus-channels<br/>(telegram, whatsapp, web)"]
        PLUGIN_SDK["roboticus-plugin-sdk<br/>(plugin registry)"]
        BROWSER["roboticus-browser<br/>(CDP automation)"]
    end

    subgraph Foundation ["Foundation"]
        CORE["roboticus-core<br/>(types, config, errors)"]
    end

    SERVER --> SCHED
    SERVER --> AGENT
    SERVER --> WALLET
    SERVER --> LLM
    SERVER --> CHANNELS
    SERVER --> PLUGIN_SDK
    SERVER --> BROWSER

    TESTS --> AGENT
    TESTS --> LLM
    TESTS --> DB

    SCHED --> AGENT
    SCHED --> WALLET
    SCHED --> DB

    AGENT --> DB
    AGENT --> LLM

    WALLET --> DB

    DB --> CORE
    LLM --> CORE
    CHANNELS --> CORE
    PLUGIN_SDK --> CORE
    BROWSER --> CORE
    WALLET --> CORE
    AGENT --> CORE

# Build
cargo build --release

# Run with config
./target/release/roboticus-server --config roboticus.toml

# Run tests
cargo test

Before cutting or shipping v0.8.0, run the canonical zero-regression gate:

just test-v080-go-live

This gate includes:

• workspace/unit/integration/regression batteries,
• bounded soak/fuzz checks,
• live CLI + dashboard/web UAT smoke checks, and
• release docs/artifact/provenance consistency checks.

Minimal roboticus.toml:

[agent]
name = "Roboticus"
id = "roboticus"

[server]
port = 18789

[models]
primary = "ollama/qwen3:8b"

Full configuration supports 24 sections: [agent], [server], [database], [models], [providers.*], [circuit_breaker], [memory], [cache], [treasury], [yield], [wallet], [a2a], [skills], [channels.*], [context], [approvals], [plugins], [browser], [daemon], [update], [tier_adapt], [personality], [obsidian], [network]. See docs/CONFIGURATION.md for all options.

Key configuration areas:

Roboticus supports two skill formats that extend agent capabilities without recompilation:

Programmatic skills with tool chains, policy overrides, and optional external scripts:

# ~/.roboticus/skills/weather.toml
[skill]
name = "weather-report"
description = "Fetch weather and format a summary"

[triggers]
keywords = ["weather", "forecast", "temperature"]

[[tool_chain]]
tool = "http_get"
args = { url = "https://api.weather.gov/points/{lat},{lon}" }

[[tool_chain]]
tool = "format_response"
args = { template = "weather_summary" }

[policy_overrides]
allow_external_http = true

[script]
path = "scripts/weather-format.py"
interpreter = "python3"

Natural-language skills injected into the LLM's system prompt:

---
name: code-review
triggers:
  keywords: ["review", "code review", "PR"]
  regex_patterns: ["review (this|my|the) (code|PR|pull request)"]
priority: 10
---

You are performing a code review. Follow these guidelines:

1. Check for correctness, edge cases, and error handling
2. Evaluate naming clarity and code organization
3. Flag security concerns (injection, auth, data exposure)
4. Suggest performance improvements where applicable
5. Be constructive — explain *why*, not just *what*

Skills are loaded from skills.skills_dir at boot with SHA-256 change detection. Hot-reload watches for file changes when skills.hot_reload = true. Script execution is sandboxed with configurable interpreter whitelist, timeout, and output size limits.

• Mutual authentication via on-chain identity (ERC-8004 registry on Base)
• Challenge-response with signed nonces + timestamps (60s window)
• ECDH ephemeral keypairs → AES-256-GCM session encryption with forward secrecy
• Per-message HMAC authentication, rate limiting, size limits
• Peer messages pass through injection defense with reduced authority
• Opacity principle: agents never expose internal memory, prompts, keys, or session history

• Authority levels: creator, self, peer, external — each with progressively restricted tool access
• Tool risk classification: Safe, Caution, Dangerous, Forbidden
• Treasury policy: per-payment caps, hourly/daily transfer limits, minimum reserve enforcement
• All decisions audit-logged to policy_decisions table

• Configurable interpreter whitelist (bash, python3, node by default)
• Environment stripping in sandbox mode (only PATH, HOME, ROBOTICUS_SESSION_ID, ROBOTICUS_AGENT_ID)
• Timeout enforcement and output truncation

85 REST routes + WebSocket upgrade for real-time events:

Detailed documentation in docs/architecture/:

Additional docs in docs/:

# Run all tests
cargo test

# Run specific crate tests
cargo test -p roboticus-core
cargo test -p roboticus-db
cargo test -p roboticus-llm
cargo test -p roboticus-agent
cargo test -p roboticus-wallet
cargo test -p roboticus-channels
cargo test -p roboticus-schedule
cargo test -p roboticus-server
cargo test -p roboticus-plugin-sdk
cargo test -p roboticus-browser
cargo test -p roboticus-tests

# Run with logging
RUST_LOG=info cargo run -- --config roboticus.toml

# Check formatting
cargo fmt --check

# Lint
cargo clippy -- -D warnings

Roboticus includes a bidirectional migration engine covering 6 data areas: configuration, personality, skills, sessions, cron jobs, and channels.

# Import from an external agent workspace
roboticus migrate import /path/to/source-root --yes

# Export to portable format
roboticus migrate export /path/to/output

# Standalone skill import with safety scanning
roboticus skills import /path/to/skill-directory

Skills pass through a safety scanner (50+ patterns across 5 categories) before import. Secrets are never stored in config files — API keys and tokens are converted to environment variable references. See ROBOTICUS.md for full details.

See CONTRIBUTING.md for branching model, PR workflow, and quality requirements.

Roboticus is open-source software licensed under Apache 2.0. There is no Roboticus token, coin, or cryptocurrency of any kind. Any token deployed using the Roboticus name is not affiliated with this project, its maintainers, or roboticus.ai. Do not purchase tokens claiming association with Roboticus.

Apache License, Version 2.0 -- see LICENSE and NOTICE.