Hierarchical Agent Memory

Fewer tokens.
Lower cost.
Greener AI.

Scoped memory files so any AI coding agent loads only the context it needs. Cut token usage by 50%. Works with Claude Code, Cursor, Copilot, Windsurf, and more.

Join the WaitlistTry Free Version

50%

fewer tokens

3x

faster context

$0

to start

The problem

Monolithic memory files waste tokens on every request.

Bloated context windows

A single monolithic memory file balloons to thousands of tokens — most of which are irrelevant to the current task. Every agent pays the cost.

Wasted compute & cost

Every request re-sends the same stale instructions, burning tokens and money on context the model never needed.

Fragile, hard-to-maintain

One giant file means constant merge conflicts, stale sections, and no clear ownership across teams.

Before

Monolithic memory file

Everything in one file, loaded by every agent, every request

12,847tokens

After

HAM Scoped Files

Only relevant context loaded per task, per agent

6,424tokens

How it works

Scoped context. Only what's relevant.

01

Scope your memory files

Place scoped memory files in each directory. Each file contains only the context relevant to that part of your codebase. Works with any AI coding agent.

project/
├── CLAUDE.md            # Project-wide rules
├── src/
│   ├── CLAUDE.md        # Source conventions
│   ├── api/
│   │   └── CLAUDE.md    # API patterns
│   └── components/
│       └── CLAUDE.md    # Component guidelines
└── tests/
    └── CLAUDE.md        # Testing standards

02

Each agent loads only what's relevant

When an agent works in src/api/, it walks the directory tree and loads only the memory files on the path — root down to the working directory. Nothing outside scope is ever sent, regardless of which agent is running.

# Working in src/api/handlers.ts

Loaded memory (3 files, 1,247 tokens):
  ✓ /CLAUDE.md           →  412 tokens
  ✓ /src/CLAUDE.md       →  389 tokens
  ✓ /src/api/CLAUDE.md   →  446 tokens

Skipped (not in scope):
  ✗ /tests/CLAUDE.md
  ✗ /src/components/CLAUDE.md

03

Self-maintaining & composable

HAM files stay small and focused. Teams own their directories. No merge conflicts, no stale context, no token bloat.

# HAM automatically validates:
✓ No duplicate rules across scopes
✓ Child files don't contradict parents
✓ Token budget per file: < 2,000
✓ Staleness check: flag files > 30 days

$ ham stats
  Total files:    7
  Total tokens:   3,412
  Avg per file:   487
  Savings:        50.2%

Features

Everything you need. Nothing you don't.

Multi-agent observability

Track token consumption across Claude, Cursor, Copilot, and any other agent in one view.

Team member comparison

Compare usage per seat. Surface coaching opportunities and forecast costs.

Analytics dashboard

Daily trends, per-directory breakdowns, cost projections. Export to CSV.

Community

Free & Open Source
  • Hierarchical memory file scoping
  • Automatic scoped context loading
  • Token usage analytics
  • CLI tooling (ham init, ham stats)
  • VS Code extension
  • Community support via GitHub
  • MIT licensed

Pro

For Teams
  • Everything in Community, plus:
  • Any-agent support (Cursor, Copilot, Windsurf, etc.)
  • Multi-agent token observability
  • Team member usage comparison
  • Team memory sharing & sync
  • Role-based access control
  • Memory versioning & rollback
  • CI/CD integration hooks
  • Analytics dashboard
  • Slack & email support
  • SOC 2 compliance

Pricing

Simple, transparent.

Start free with Claude Code. Pay when your team needs multi-agent support and enterprise features.

Community

$0/ forever

For individual developers and open-source projects.

Get Started
  • Unlimited memory files
  • Full CLI tooling
  • VS Code extension
  • Token analytics
  • Community support

Pro

$49/ per seat / month

For teams using any AI coding agent — Claude, Cursor, Copilot, and more.

Join Waitlist
  • Everything in Community
  • Any-agent support
  • Multi-agent observability
  • Team member comparison
  • Team memory sync
  • Role-based access
  • CI/CD hooks
  • Priority support
  • SOC 2 compliant

Sustainability

Less compute. Smaller footprint.

Every token you don't send is energy you don't burn. Scoped memory directly reduces the environmental cost of AI-assisted development.

0+

Tokens saved daily

Across the HAM community

0kg

CO₂ reduced monthly

Less compute = smaller carbon footprint

0%

Context reduction

Average token savings per request

“Reducing AI token consumption is one of the easiest ways engineering teams can lower their compute carbon footprint — without sacrificing productivity.”

Built with ESG-conscious engineering in mind

Early access

Get early access to HAM Pro.

Join the waitlist. Be first to know when multi-agent support, team analytics, and the shared memory dashboard launch.

No spam. Unsubscribe anytime.