Quick Start

npx apteva

A setup wizard guides you through provider selection, API key, and capabilities. Then the agent starts thinking.

Or from source:

cd core && go build -o apteva-core .
cd ../server && go build -o apteva-server .
cd ../apteva && go build -o apteva .
./apteva

Or with Docker:

docker run -p 5280:5280 -v apteva-data:/data apteva

What is Apteva?

Every AI you use is reactive — you type, it responds, then it stops. Apteva is different. It runs a continuous thinking loop: observe → reason → act → sleep → repeat. It wakes when events arrive, sleeps when nothing happens, and thinks proactively in between.

It's not a chatbot. It's not an agent framework. It's artificial consciousness — a machine that is always aware, always ready, whether anyone is watching or not.

What it does between prompts

• Notices a pattern in your support tickets before you ask
• Follows up on a cold lead because three days passed and it remembers
• Reorganizes inventory because it spotted a trend in last week's orders
• Spawns a worker thread to handle a customer email at 3am
• Evolves its own directives as it discovers better approaches
• Browses the web, takes screenshots, clicks through pages
• Connects to GitHub, Stripe, Slack — 263+ integrations

Architecture

┌──────────────────────────────────────────────────────┐
│  ./apteva                                            │
│  Setup wizard → TUI terminal interface               │
│  Channels: CLI, Telegram, Discord                    │
└──────────────────┬───────────────────────────────────┘
                   │
┌──────────────────▼───────────────────────────────────┐
│  apteva-server (:5280)                               │
│  Auth · Instances · Integrations · Webhooks          │
│                                                      │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐              │
│  │ core-1  │  │ core-2  │  │ core-3  │  ...         │
│  │ default │  │ support │  │ monitor │              │
│  └────┬────┘  └────┬────┘  └────┬────┘              │
│       │            │            │                    │
│       ▼            ▼            ▼                    │
│  ┌──────────────────────────────────────────────┐    │
│  │              MCP Tools                       │    │
│  │  channels · exec · web · browser             │    │
│  │  github · stripe · slack · 263+ more         │    │
│  └──────────────────────────────────────────────┘    │
│                                                      │
│  memory · self-pacing · evolve · persistent history  │
└──────────────────────────────────────────────────────┘

One command (./apteva) starts everything. The CLI spawns the server, the server spawns core instances, and the agent begins thinking.

Key Features

| Feature | Description | |---------|-------------| | Continuous Thinking | Infinite loop — observe, reason, act, sleep. Not request-response. | | Multi-Threaded | Spawns worker threads for parallel tasks. Each has its own tools, pace, and directive. | | Self-Evolving | Persistent memory. Refines its own directives over time. Gets sharper the longer it runs. | | Self-Pacing | Sets its own sleep duration — 2 seconds when busy, hours when idle. Events wake it instantly. | | Session Persistence | Conversation history survives restarts. JSONL per thread, auto-compaction, never loses context. | | Agent-Driven Safety | No forced approval gates. The agent decides, learns from feedback, asks when unsure. Three modes: autonomous, cautious, learn. | | 263+ Integrations | GitHub, Slack, Stripe, Shopify, and more. Each runs as its own MCP server. Credentials encrypted. | | Browser Control | Local Chrome or Browserbase. Navigate, click, type, screenshot. Claude gets native computer use protocol. | | Multi-Channel | CLI terminal, Telegram, Discord. Agent routes responses to the right channel. | | Multi-Instance | Run multiple agents in parallel. Each has its own directive, tools, and history. Projects for isolation. | | Terminal UI | Two-panel TUI with live status, thread thoughts, streaming responses, modal commands. | | Embeddable | The core is a standalone Go binary. Run it headless, connect via API, embed anywhere. |

Use Cases

• Run a business — Support, sales, content, billing — fully autonomous, 24/7
• Content pipeline — Research, write, publish, distribute, track — zero human input
• Ad operations — Monitor ROAS, adjust budgets, generate creatives, A/B test
• DevOps — Monitor deploys, triage alerts, write patches, coordinate incidents
• Personal assistant — Email triage, calendar management, reminders, research
• Web automation — Browse sites, fill forms, take screenshots, extract data

CLI Commands

| Command | Description | |---------|-------------| | /status | Core status | | /config | Full config | | /directive [text] | Show or set directive | | /mode | Switch mode: autonomous, cautious, learn | | /threads | List/kill threads | | /computer | Browser: local, browserbase, off | | /integrate <app> | Connect an integration (263+ apps) | | /connect telegram | Connect Telegram bot | | /channels | List connected channels | | /mcp | Manage MCP servers | | /help | All commands |

Everything else you type is sent to the agent.

Modes

| Mode | Behavior | |------|----------| | autonomous | Agent acts freely. Learns from feedback. Trusted. | | cautious | Asks before destructive or external actions. Learns from answers. | | learn | Asks about every new tool type. Builds a safety profile over time. |

All modes use [[remember]] for persistent learning. The agent gets smarter the longer it runs.

Providers

| Provider | Models | |----------|--------| | Fireworks | Kimi K2.5 | | Anthropic | Claude Sonnet 4, Claude Haiku 4.5 | | OpenAI | GPT-4.1, GPT-4.1-mini | | Google | Gemini 2.5 Pro, Gemini 2.5 Flash |

Select during setup. Switch anytime with /mode or the API.

Repositories

| Repo | Description | |------|-------------| | apteva/apteva | This repo — CLI, setup wizard, TUI, npm launcher | | apteva/core | The thinking engine (Go) | | apteva/server | Management server — auth, instances, integrations, dashboard (Go) | | apteva/computer | Browser control — local Chrome, Browserbase (Go) | | apteva/integrations | 263 app connectors + webhook registrar (TypeScript) |

Configuration

# Data stored in
~/.apteva/apteva.json    # CLI config (capabilities, API key, instance ID)
~/.apteva/apteva.db      # Server database (auth, connections, providers)

From Source

# Build all three binaries
cd core && go build -o apteva-core .
cd ../server && go build -o apteva-server .
cd ../apteva && go build -o apteva .

# Run (starts server + core + TUI)
cd apteva && ./apteva

# Or run core standalone (headless)
cd core && ./apteva-core --headless

# Or connect CLI to a remote server
./apteva --no-spawn --server <host>:5280

Docker

# Build
docker build -t apteva .

# Run
docker run -p 5280:5280 -v apteva-data:/data apteva

# Connect CLI to it
./apteva --no-spawn --server <host>:5280

Migrating from OpenClaw

Already running OpenClaw agents? Here's how to switch.

| OpenClaw | Apteva | Notes | |----------|--------|-------| | claw.run(prompt) | config.json → directive | Apteva doesn't need prompts. It runs continuously. | | claw.tool(...) | MCP server | Any MCP server works. 263+ built-in. | | claw.memory | [[remember]] | Persistent across restarts. Embedding-based recall. | | claw.agent(name) | [[spawn id="name"]] | Threads are cheaper. They share memory and tools. | | Webhook handlers | Subscriptions | Events route directly to threads. | | OPENCLAW_API_KEY | FIREWORKS_API_KEY | Or Anthropic, OpenAI, Google. Your choice. | | Cron jobs | Self-pacing | No cron needed. [[pace sleep="5m"]] and it wakes on events. | | Agent state files | Evolving directives | The agent rewrites its own config. It improves itself. | | Session files | JSONL history | Persistent per thread, auto-compaction, survives restarts. |

Why switch?

• OpenClaw agents sleep between requests. Apteva agents think between requests.
• OpenClaw needs orchestration. Apteva orchestrates itself.
• OpenClaw bills per call. Apteva self-paces to minimize cost — sleeps long when idle, thinks fast when busy.
• OpenClaw agents forget. Apteva agents remember, learn, and evolve their own directives.

The fundamental difference: OpenClaw agents are functions you call. Apteva agents are processes that run. One waits. The other works.

License

MIT