Skip to main content

What Can You Build?

Bossa gives your agents a persistent filesystem. Here are concrete examples aligned with modern agent patterns: coding assistants, multi-agent collaboration, and context engineering.

1. Coding Assistant

Your coding agent remembers project context, tech stack, and conventions across sessions. No more re-explaining your codebase.

File Structure

/projects
  /my-app
    tech-stack.json
    conventions.md
    architecture.md
    decisions/

Example: Store and Retrieve Project Context

# Session 1: Capture project context
bossa files write /projects/my-app/tech-stack.json '{"framework": "FastAPI", "db": "Postgres", "style": "async"}'
bossa files write /projects/my-app/conventions.md '# Code style: type hints, pydantic models, async I/O'

# Session 2: Agent pulls context on demand
bossa files read /projects/my-app/conventions.md
# Agent: "I'll follow your FastAPI + async patterns."

Example: Search Across Decisions

bossa files grep "auth" --path /projects/my-app/decisions

2. Multi-Agent Team

Multiple agents share one workspace. Agent A writes context; Agent B reads it. Handoffs, shared state, no duplicate work.

File Structure

/agents
  /planner
    current-task.md
    next-steps.md
  /coder
    implementation-notes.md
  /reviewer
    feedback.md

Example: Agent Handoff

# Planner agent writes task for coder
bossa files write /agents/planner/current-task.md 'Implement auth middleware. Use JWT. See /agents/planner/next-steps.md'

# Coder agent reads and executes
bossa files read /agents/planner/current-task.md
bossa files read /agents/planner/next-steps.md
# Coder: implements, then writes notes for reviewer
bossa files write /agents/coder/implementation-notes.md 'Auth done. Edge cases: token expiry, refresh flow.'

Example: Shared Discovery

bossa files ls /agents
bossa files grep "auth" --path /agents

3. Context Engineering (Dynamic Context Discovery)

Instead of loading everything into context, agents discover what’s relevant with ls and grep, then pull only what they need. Fewer tokens, better answers. Same pattern Cursor and LangChain use for dynamic context.

File Structure

/context
  /user
    preferences.json
  /projects
    /alpha
      summary.md
      decisions/
    /beta
      summary.md

Example: Discover Then Read

# Agent discovers what exists
bossa files ls /context/projects

# Agent searches for relevant context
bossa files grep "database" --path /context

# Agent reads only what it needs
bossa files read /context/projects/alpha/summary.md

Example: Avoid Token Bloat

# Don't load everything—grep first, read second
bossa files grep "migration" --path /context/projects/alpha
# Then: bossa files read /context/projects/alpha/decisions/migration-2026-03.md

4. Session Memory (Learn Once, Remember Forever)

The simplest pattern: learn something in session 1, use it in session 2.
# Session 1: Learn something
bossa files write /learned/facts.json '{"user_likes": "vim", "preferred_lang": "Python"}'

# Session 2: Remember it
bossa files read /learned/facts.json
# Agent: "I remember you like vim!"

5. Research Agent

Build up a knowledge base over weeks. Organize sources, findings, and notes by topic. Search across everything.

File Structure

/research
  /ai-trends-2026
    sources.md
    findings.json
  /context-engineering
    papers.md
    notes/

Example: Store and Search Research

bossa files write /research/context-engineering/findings.json '{
  "key_concepts": ["dynamic context", "pull-on-demand", "ls/grep pattern"],
  "sources": ["cursor-blog", "langchain-blog"]
}'
bossa files grep "dynamic context" --path /research

6. Customer Support Agent

Never ask a customer the same question twice. Store account info, ticket history, and interaction context.

File Structure

/customers
  /acme-corp
    account-info.json
    support-tickets/
    interaction-history.md

Example: Store and Search Customer Context

bossa files write /customers/acme-corp/account-info.json '{
  "company": "Acme Corp",
  "plan": "enterprise",
  "notes": "Prefers email over chat"
}'
bossa files grep "enterprise" --path /customers

Next Steps

Last modified on March 13, 2026