Hierarchical Agent Memory
Save 50% on AI token costs.
Fewer tokens. Lower cost. Greener AI.
50%
lower token costs
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. At scale, this is the fastest-growing line item on your API bill.
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.
Before
12,847 tokens
After
12,847 tokens
agent → handlers
01
Your agent starts working
Your AI agent opens a file deep in your codebase — say src/api/handlers.ts. Without HAM, it would load one giant memory file containing every rule for every directory, whether relevant or not.
02
HAM traces the path
HAM walks from your project root down to the working directory, collecting only the scoped memory files along that path. Rules for /tests or /components never load when the agent is working in /api.
03
You only pay for what matters
Your agent gets exactly the context it needs — nothing more. Token usage drops, responses get faster, and your API bill shrinks. The same work, half the cost.
Want to see how it works under the hood? Read the docs
Why HAM
Built for how you actually work.
Most teams start with one big memory file and hope for the best. It works — until it doesn't. Context grows, agents slow down, and your token bill doubles before anyone notices.
HAM replaces that pattern with scoped, hierarchical memory files that mirror your directory structure. Each agent loads only the context it needs, from root to working directory. No wasted tokens, no stale instructions, no bloat.
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
For individual developers and open-source projects.
- Unlimited memory files
- Full CLI tooling
- VS Code extension
- Token analytics
- Community support
Pro
For teams using any AI coding agent — Claude, Cursor, Copilot, and more.
- 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
Your company doesn't know this yet. You do now.
Every token your AI agent processes costs money, burns electricity, and emits carbon. Same action, three ledgers. When your API bill goes up, your carbon footprint goes up with it.
Your finance team is wondering why cloud costs keep climbing. Your sustainability team is filing last year's numbers with no visibility into AI workloads. You're the only person at your company who sees both sides.
HAM cuts token usage by 50%. Lower API costs. Faster agents. Less compute carbon. One tool that gives you an answer for every room you walk into.
0%
Fewer tokens
One metric. Three wins.
0 in 11
Companies tracking AI's environmental impact
Be the one who brings this up.
0x
Carbon gap between efficient and wasteful prompts
Context scoping matters.
“When your CFO asks why API costs are up and your sustainability team can't account for AI in their reporting, you'll be the one with the answer.”