Getting Started with ABF

This guide walks you through installing ABF, creating your first project, running your agents, and understanding what happens under the hood. It should take under 5 minutes.

---

Prerequisites

• Node.js 20 or later -- Check with node --version
• pnpm 10 or later -- Install with npm install -g pnpm if needed
• An API key for at least one LLM provider (Anthropic, OpenAI) or a running Ollama instance for local-only operation

---

1. Install the CLI

npm install -g @abf/cli

Verify:

abf --version
# 1.0.0

Alternatively, use npx without installing globally:

npx @abf/cli init --template solo-founder --name my-business

---

2. Create a Project

ABF provides three paths to create a project. Pick the one that fits your situation.

Path A: Start from a template (fastest)

abf init --template solo-founder --name my-business
cd my-business

This creates three agents (Compass the executive assistant, Scout the researcher, and Scribe the content writer) organized into a single team. You should see output like:

Created my-business/abf.config.yaml
Created my-business/agents/compass.agent.yaml
Created my-business/agents/scout.agent.yaml
Created my-business/agents/scribe.agent.yaml
Created my-business/teams/founders.team.yaml
Created my-business/memory/decisions.md
Created my-business/knowledge/company.md
Created my-business/knowledge/brand-voice.md
Created my-business/docker-compose.yml
Created my-business/README.md

Project "my-business" created with solo-founder template.

Available templates:

Template	Command	Agents	Description
Solo Founder	--template solo-founder	3	Executive assistant, researcher, writer
SaaS Startup	--template saas	5	Product + go-to-market teams
Marketing Agency	--template marketing-agency	4	Director, strategist, copywriter, analyst

Path B: Start from a business plan

If you have a business plan, pitch deck, or company description document (.docx, .pdf, .txt, or .md):

abf init --seed ./my-business-plan.md

ABF parses the document, sends it to an LLM for analysis, and generates a complete project with agents, teams, knowledge files, and workflows tailored to your specific business. See the full Seed-to-Company Guide for details.

Path C: Start from the setup wizard

For a fully visual experience:

abf setup

This opens the Dashboard setup wizard in your browser, where you can choose between answering interview questions about your business idea, uploading a document, or picking a template -- all through a guided UI.

---

3. Configure an LLM Provider

ABF needs access to at least one LLM to power your agents. The default templates use Anthropic (Claude).

Anthropic (recommended)

abf auth anthropic
# Enter your Anthropic API key: sk-ant-...
# Credential stored securely.

OpenAI

abf auth openai

After adding the key, update each agent YAML to use OpenAI:

provider: openai
model: gpt-4o

Ollama (local, no API key)

Start Ollama on your machine, then:

abf auth ollama

Update each agent YAML:

provider: ollama
model: llama3.2

Any OpenAI-compatible provider

ABF supports any provider with an OpenAI-compatible API (Groq, Together, Fireworks, local vLLM, etc.). Configure in abf.config.yaml:

providers:
  - slug: groq
    name: Groq
    type: openai-compat
    base_url: https://api.groq.com/openai/v1
    api_key_env: GROQ_API_KEY
    models:
      - llama-3.3-70b-versatile

Then reference it in agent YAML:

provider: groq
model: llama-3.3-70b-versatile

Verify your setup

abf auth --list
# Configured providers:
#   anthropic  ok

---

4. Start the Runtime

abf dev

Expected output:

ABF runtime starting...
  Loaded 3 agents: compass, scout, scribe
  Loaded 1 team: founders
  Gateway listening on http://localhost:3000
  Scheduler started (2 cron triggers registered)

The runtime is now running with all five components:

• Scheduler -- Fires cron and heartbeat triggers
• Dispatcher -- Handles agent activations, spawns sessions
• Session Manager -- Loads context, calls LLM, runs tools, writes memory
• Bus -- Routes messages between agents
• Gateway -- Serves the REST API and Dashboard on port 3000

Leave this terminal running and open a new one for the next steps.

---

5. Open the Dashboard

Navigate to:

http://localhost:3000

You will see the ABF Dashboard with:

• Overview -- System status, number of active agents, recent sessions
• Agents -- Your three agents with status indicators
• Teams -- The "founders" team with Compass as orchestrator

Click on any agent to see its details: charter, configuration, KPIs, and a form to send tasks to its inbox.

---

6. Run Your First Agent

From the CLI

In a second terminal:

abf run compass --task daily_briefing

This triggers Compass to execute its daily_briefing task. You will see the session output in your terminal, including any tool calls, LLM responses, and the final output.

From the Dashboard

1. Go to the Agents page
2. Click on Compass
3. Use the "Send Task to Inbox" form
4. Type a task like "Give me a briefing on today's priorities"
5. Click Send

The agent picks up the task on its next activation.

---

7. What Just Happened

When an agent runs a session, ABF executes an 8-step lifecycle:

1. Load Context -- Reads the agent's charter, history, team decisions, and knowledge files
2. Build Prompt -- Assembles the system prompt with date, KPIs, pending messages, and inbox items
3. Call LLM -- Sends the prompt to the configured provider (Anthropic, OpenAI, Ollama)
4. Tool Loop -- Executes any tool calls the LLM requests, returns results, repeats
5. Process Outputs -- Routes any inter-agent messages to the bus
6. Write Memory -- Appends learnings to the agent's history file
7. Check Escalations -- Routes to human or orchestrator if escalation rules trigger
8. Report -- Updates KPIs, logs cost, closes the session

After a session completes, you can find the results in several places:

Session logs:

abf logs --agent compass

Agent history (accumulated learnings):

memory/agents/compass/history.md

Session outputs (readable by teammate agents):

outputs/compass/2026-03-01T10-00-00.md

Team decisions:

memory/decisions.md

---

8. Customize an Agent

Open an agent definition to modify its behavior:

$EDITOR agents/scout.agent.yaml

Change the model

provider: openai
model: gpt-4o
temperature: 0.3

Adjust triggers

triggers:
  - type: cron
    schedule: '0 */4 * * *'     # Every 4 hours
    task: research_scan
  - type: message
    from: compass               # Triggered when Compass sends a message
    task: on_demand_research
  - type: manual
    task: research_scan

Tighten behavioral bounds

behavioral_bounds:
  allowed_actions: [read_data, write_report]
  forbidden_actions: [delete_data, modify_billing, access_credentials]
  max_cost_per_session: \$0.50
  requires_approval: [publish_content, send_client_email]

Edit the charter

The charter field is the agent's identity and instructions. Change how the agent thinks and responds by editing this field.

After making changes, restart the runtime (Ctrl+C and abf dev again) to pick up the new configuration.

---

9. Add a New Agent

Scaffold from a built-in archetype:

abf agent add --name analyst --archetype analyst --team founders

This creates agents/analyst.agent.yaml with sensible defaults for an analyst: lower temperature, analytical tools, and a starter charter.

Available archetypes:

Archetype	Temperature	Default Tools	Description
researcher	0.3	web-search, knowledge-search	Deep research and information gathering
writer	0.7	knowledge-search, image-render	Content creation and drafting
orchestrator	0.2	send-message, knowledge-search	Team coordination and task routing
analyst	0.2	database-query, knowledge-search	Data analysis and reporting
customer-support	0.4	send-message, knowledge-search, database-query, email-send, privacy-ops	Customer interaction and issue resolution
developer	0.3	knowledge-search, github-ci, app-generate, app-deploy, backend-provision, code-generate	Code, PRs, deployments, and technical solutions
marketer	0.6	web-search, knowledge-search, send-message, email-send, image-render, social-publish	Marketing strategy, campaigns, and growth
finance	0.1	database-query, knowledge-search, stripe-billing, privacy-ops	Financial analysis, billing, and reporting
monitor	0.1	web-search, knowledge-search, send-message	Watching for changes and alerting
generalist	0.4	knowledge-search	General-purpose agent

---

10. Next Steps

You now have a running ABF project. Here is where to go next:

Goal	Guide
Understand ABF's mental model	Concepts
Generate a custom agent team from a business plan	Seed-to-Company Guide
Deploy to production	Self-Hosting Guide
Build multi-agent workflows	Add a workflow: abf workflow add --template sequential-pipeline --name my-flow
Set up a business database	Add datastore config to abf.config.yaml, then abf migrate
Write a custom tool	Create tools/my-tool.tool.yaml + tools/my-tool.tool.js
Add shared knowledge	Drop .md files into the knowledge/ directory
Monitor external URLs	Create monitors/my-monitor.monitor.yaml
Explore the API	API Reference

---

Troubleshooting

"Command not found: abf"

The CLI is not installed globally. Either install it:

npm install -g @abf/cli

Or use npx:

npx @abf/cli dev

"No provider configured"

You need at least one LLM provider. Run abf auth anthropic (or openai / ollama) and enter your API key.

Agent sessions fail with "model not found"

The agent YAML references a model that your configured provider does not support. Check the provider and model fields in the agent YAML match your configured provider.

Port 3000 is already in use

Start on a different port:

abf dev --port 3001

Dashboard shows no agents

Make sure you are in the project directory (the one containing abf.config.yaml) when running abf dev.