snsagent is a personal, single-user AI coding agent for the terminal. Bring your own API key (BYOK), talk to the agent, and it configures itself — MCP servers, memory backends, model switching — through conversation. No forced subscription; the ~120 MB binary is yours to run anywhere.

Table of Contents

• Features
• Quick Start
• Installation
• Termux (Android)
• Configuration
• Telegram
• Architecture
• Documentation
• Development
• Security
• License

Features

Status legend: VERIFIED = a real integration path was demonstrated (not just a green unit test). PARTIAL = implemented and partly tested, but not deep-audited end-to-end. UNTESTED = implemented, no committed test. BROKEN = dead/unreachable code.

| Feature | Status | Evidence | |---------|--------|----------| | 30 built-in tools | VERIFIED | src/tools/builtin-names.ts (all real implementations) | | 62 slash commands (+ 4 aliases) | VERIFIED | src/slash-commands/builtin-registry.ts | | Multi-provider LLM (OpenAI, Anthropic, Ollama, custom) | VERIFIED | src/modes/setup-wizard/ + provider registry | | BYOK setup wizard | VERIFIED | src/modes/setup-wizard/ + tests + screenshots | | Memory — mnemopi (default recommended) | VERIFIED | src/memory-backend/__tests__/memory-integration.test.ts (30 tests: retain → persist → recall → inject) | | Memory — hindsight (remote service) | PARTIAL | src/hindsight/; needs hindsight.apiUrl service — docs/memory.md | | Memory — mem0 / lcm / local | PARTIAL | manual save/search; no auto-recall/injection — docs/memory.md | | Memory — mnemosyne | BROKEN | migrated to mnemopi at config load; dead enum value — docs/memory.md | | TBM (token budget manager) | VERIFIED (integrated) | src/tbm/session-hooks.ts wired into the turn lifecycle; default OFF. No savings claim without the measured benchmark — docs/tbm.md | | Telegram bot | PARTIAL | src/adapters/telegram/; auth via opt-in SNS_TELEGRAM_ALLOWED_USERS — docs/telegram.md | | MCP integration | PARTIAL | src/mcp/ (22 files); /mcp surface verified — docs/mcp.md | | Plan mode | PARTIAL | src/plan-mode/ — docs/plan-mode.md | | Goal mode | PARTIAL | src/goals/ (token budget + lifecycle) — docs/goals.md | | Subagents / multi-agent | PARTIAL | src/task/, src/agents/ (consensus/critic/best-of-N) — docs/subagents.md | | Advisor (second-model review) | VERIFIED | src/advisor/ + advisor.test.ts — docs/advisor.md | | Cron scheduler | PARTIAL | src/cron/ + parser tests — docs/cron.md | | Browser automation (Puppeteer) | PARTIAL | src/tools/browser/ — docs/browser.md | | SSH remote execution | PARTIAL | src/tools/ssh.ts | | Eval backends (Python/JS/Ruby/Julia) | VERIFIED | src/eval/ + eval tests | | Context compaction | PARTIAL | src/session/ (multiple strategies) — docs/compaction.md | | Plugins / skills | UNTESTED | src/extensibility/ (implemented, no committed test) — docs/extensibility.md | | Collaborative sessions | UNTESTED | src/collab/ (implemented, no committed test) — docs/collab.md | | LSP integration | PARTIAL | src/lsp/ | | Text-to-speech / STT | PARTIAL | src/tts/, src/stt/ | | Todo system | PARTIAL | src/ todo helpers |

Full findings: AUDIT-REPORT.md, AUDIT-BASELINE.md, docs/security-model.md.

Tools and commands. 30 built-in tools, 62 slash commands:

Settings and model picker. Configure everything through /settings or conversation:

Memory. mnemopi backend (local SQLite + vector + graph):

TBM. Token Budget Manager dashboard (integrated, default OFF):

MCP. Model Context Protocol integration surface:

Goal mode, cron, subagents, browser, compaction:

Stats and error handling:

Collab sessions:

Compiled binary note. ./bin/snsagent-linux-x64 (and the npm-installed binary) runs in "JS-only mode" - the native pty/grep/shell addon is not embedded, so the interactive TUI does not render there. The source-run path (bun run src/cli/entry.ts) is the supported path for the interactive UI. See docs/troubleshooting.md.

Quick Start

1. Run

snsagent

First launch opens the setup wizard - choose a provider (or custom Base URL), enter an API key, pick a model, and start chatting.

Once configured, you land in the main TUI:

Type /help anytime to see all slash commands:

2. Use it

> what files are in the current directory?
> search the web for "bun runtime benchmarks"
> create a TypeScript module that parses CSV files
> refactor src/utils/edit-mode.ts to use async/await

3. Configure through conversation

> add MCP filesystem for /home/user/projects
> switch to anthropic with claude-sonnet
> load coding skill

Or set the key directly via environment — the agent picks it up on launch:

export OPENAI_API_KEY="your-key-here"   # or ANTHROPIC_API_KEY, etc.
snsagent

Installation

npm (recommended for most users)

npm install -g @sns-myagent/cli
snsagent

The package downloads the platform binary during install. Requires Node.js 18+; works on Linux, macOS, Windows, and WSL.

One-liner installer

curl -fsSL https://raw.githubusercontent.com/Reihantt6/sns-myagent/main/install.sh | bash

• Linux, macOS, WSL: curl … install.sh | bash
• Windows PowerShell: irm https://raw.githubusercontent.com/Reihantt6/sns-myagent/main/install.ps1 | iex
• Android/Termux: see Termux (Android)

From source

git clone https://github.com/Reihantt6/sns-myagent.git
cd sns-myagent
bun install
bun run src/cli/entry.ts

Requires Bun >= 1.3.14. To build the standalone Linux x64 binary:

bun run build
./bin/snsagent-linux-x64

Verify

snsagent --version   # -> snsagent 0.3.9
snsagent --help

Termux (Android)

Full guide: docs/termux.md. The npm/prebuilt binaries are glibc and will not run on Android — use the source build.

# 1. Install Termux from F-Droid (NOT Play Store)
pkg update && pkg upgrade -y
pkg install -y git nodejs-lts
# 2. Install Bun
curl -fsSL https://bun.sh/install | bash
source ~/.bashrc
# 3. Clone + run from source
git clone https://github.com/Reihantt6/sns-myagent.git
cd sns-myagent
bun install
bun run src/cli/entry.ts

Configuration

Interactive settings live at ~/.omp/agent/config.yml (providers at ~/.omp/agent/models.yml). Edit them through /settings or through conversation; the schema defaults are safe (memory backend defaults to off, TBM defaults to off).

Key settings

{
  // Model selection
  "model.*": "…",

  // Feature toggles
  "advisor.enabled": true,
  "bash.enabled": true,
  "browser.enabled": true,
  "checkpoint.enabled": true,
  "compaction.enabled": true,
  "goal.enabled": true,
  "todo.enabled": true,

  // Memory (default "off"; "mnemopi" is the fully-integrated local backend)
  "memory.backend": "off",   // off | mnemopi | hindsight | mem0 | lcm | local | mnemosyne

  // TBM (default OFF — no behavior change until opted in)
  "tbm.enabled": false,

  // Eval backends
  "eval.py": true, "eval.js": true, "eval.rb": true, "eval.jl": true
}

Full schema: docs/configuration.md and src/config/settings-schema.ts.

Where things are stored

| What | Location | |------|----------| | Interactive config | ~/.omp/agent/config.yml | | Provider models | ~/.omp/agent/models.yml | | Memory & sessions | ~/.omp/agent/ (mnemopi data, sessions, state) | | Daemon secrets (systemd) | /etc/snsagent/secrets.env (chmod 600) |

Telegram

Run the agent as a Telegram bot. Built on grammY.

export SNS_TELEGRAM_BOT_TOKEN="your-bot-token-here"
snsagent          # bot auto-starts polling

Or explicitly: snsagent telegram. Disable auto-start with SNS_TELEGRAM_AUTOSTART=0.

Authorization is opt-in and important. Without it the bot executes agent actions for anyone who can message it. Restrict to your own numeric user id:

export SNS_TELEGRAM_ALLOWED_USERS="123456789"

See docs/security-model.md for the full authorization boundary (including the autoApprove: true caveat).

Architecture

sns-myagent/
├── src/
│   ├── cli/                  # CLI entry point + commands
│   ├── config/               # Settings schema, loader, defaults
│   ├── tools/                # 30 built-in tools
│   ├── modes/                # Interactive mode, TUI, controllers
│   ├── session/              # Session management + agent loop
│   ├── mcp/                  # MCP integration
│   ├── memory-backend/       # Memory backend resolver
│   ├── mnemopi/              # mnemopi backend (SQLite + vector + graph)
│   ├── hindsight/            # hindsight backend (remote memory)
│   ├── tbm/                  # Token Budget Manager (integrated, default OFF)
│   ├── task/ agents/         # Subagent delegation / multi-agent
│   ├── plan-mode/ goals/     # Plan mode / goal mode
│   ├── advisor/              # Second-model review
│   ├── cron/                 # Cron scheduler
│   ├── eval/                 # Eval runtimes (Python/JS/Ruby/Julia)
│   ├── extensibility/        # Skills, plugins, marketplace
│   ├── collab/               # Collaborative sessions
│   ├── adapters/telegram/    # Telegram bot adapter
│   ├── tts/ stt/             # Speech-to-text / text-to-speech
│   ├── lsp/ dap/             # LSP + Debug Adapter Protocol
│   └── utils/                # Utilities
├── docs/                     # Documentation (+ screenshots/)
├── bin/                      # Prebuilt binaries (gitignored)
├── scripts/                  # Build + dev scripts
├── install.sh / install.ps1  # Installers
└── package.json

Documentation

| Guide | Covers | |-------|--------| | Architecture | Agent-loop integration points (memory/TBM/advisor/tools) | | Installation | Install options per platform | | Configuration | Config files, providers, models | | Memory | Memory backends and how to switch | | MCP | Model Context Protocol servers | | Termux (Android) | Running on Android/Termux | | Terminal UI | The interactive TUI | | TBM | Token budget manager (integrated, default OFF) | | Telegram | Telegram bot + authorization boundary | | Goals | Autonomous objective mode | | Plan mode | Plan-before-execute workflow | | Subagents | Task delegation + async background jobs | | Browser | Puppeteer browser automation | | Cron | Scheduled jobs (prompt/shell/skill) | | Compaction | Context-window compaction strategies | | Extensibility | Plugins, skills, custom tools | | Collab | Live shared sessions | | Advisor | Second-model turn review | | Security model | Authorization + attack surface | | Upstream comparison | Lineage vs oh-my-pi | | Syarat & Ketentuan | Terms of use (Indonesian) | | Development | Custom node_modules + install mechanism | | Troubleshooting | Common issues and fixes | | FAQ | Frequently asked questions |

Development

bun install              # Install dependencies (runs custom postinstall — see docs/development.md)
bun run build            # Build binary to bin/
bun run src/cli/entry.ts # Run from source
bun test                 # Run tests
bunx biome lint src test # Lint
bunx tsc -p tsconfig.json --noEmit  # Typecheck

See docs/development.md for the custom node_modules install mechanism (binary download + pi-natives JS-only patch) and dependency drift.

Security

snsagent executes code (bash, eval, SSH, browser, MCP tools) on your behalf. Review the authorization boundary and tool-approval behavior in docs/security-model.md. See SECURITY.md for vulnerability reporting.

License

MIT license. snsagent is a rebranded fork of Pi Agent with implementation lineage from the @oh-my-pi/* packages (see docs/upstream.md).