Indexes What You Already Have
Point Agent Memory at folders of Markdown, source code, runbooks, decision logs, JSON, YAML and more. It chunks, embeds and stores them locally — no manual prep, no upload step.
Private Local Memory · Windows · Mac · Linux
One install. One memory.
Every agent on your machine.
Agent Memory turns your PC into a private memory layer for any AI assistant — indexing your folders, repos, runbooks and notes into a local semantic store, ready to use through a single local API. Eighteen built-in skills cover Organising, Design and Marketing out of the box. Always your data, always your call.
- 1‑Click Install
- 100% Local Indexing
- 18 Built-In Skills
- Connects Any Agent
http://127.0.0.1:8765 · Ready
Search a project
“What did we decide about staging last week?”
Searching local memory…
# /search · 0.42s · 8 chunks
1. docs/HANDOVER.md score 0.81
"Use the staging domain first.
Cutover is reversible."
2. PR-notes/2026-04-19.md score 0.74
"Owner: Marc.
Smoke tests, then prod."
3. decision-logs/staging.md score 0.69
"Rollback: switch DNS back."Connect any agent
One local API, every assistant on the same page.
- Claude Desktop
- ChatGPT Desktop
- Cursor
- Continue.dev
- n8n & Zapier
- Your own scripts
POST /search → profilePrompt + ranked snippets
$29
One-time purchase. Lifetime licence. No subscription required.
18skills
Organising · Design · Marketing — six recipes per category, ready to run.
100% local
Indexing, embeddings and search all run on your machine. Nothing phones home.
Built for continuity
Give every assistant the context it should have had already.
Most agents start every conversation with amnesia. Agent Memory keeps your project knowledge, decisions and working preferences once — locally — and serves them through a single API any assistant on your machine can call.
Semantic Search, Locally
Queries are answered by a local MiniLM embedding model and a cosine-similarity index on disk. No hosted vector database, no per-query bill, no rate limits.
Remembers Your Working Style
Save your About Me, How I Like to Work and Very Important guidance once. Agent Memory returns it as a profilePrompt before every search result, so assistants stay consistent.
One Local API, Any Agent
A small Express API on 127.0.0.1:8765 exposes /search, /ingest, /documents, /profile and /stats. Connect Claude Desktop, ChatGPT Desktop, Cursor, n8n, or your own scripts.
Your Files Never Leave
The vector index is a JSON file under your user data folder. The API binds to 127.0.0.1, not the public network. Nothing is sent off-device unless you explicitly choose to.
As Open As You Want
Add manual notes by hand, swap embedding models, inspect the JSON store, edit a skill, or rebuild the index from scratch. Friendly defaults for newcomers, full control for power users.
The Skills Library — built in
Eighteen ready-made recipes for the work you already do.
Skills are short operating procedures stored alongside your local memory. Ask any connected assistant to “run a Project Briefing,” “draft a Launch Brief,” or “do a Reading Order Audit,” and the skill provides the recipe while Agent Memory provides the project context. Edit them, write your own, or share a folder of skills across machines.
Organising
ORG · 6Skills that turn scattered notes, repos and decisions into something you can actually act on.
Project Briefing
Search the local index for a project and produce a one-page brief with cited file paths.
Reads docs/, runbooks/→Produces 1‑page brief
Decision Recap
Pull recent decision-log entries and summarise what changed, who decided, and what is reversible.
Reads decision-logs/→Produces recap with owners
Runbook Writer
Turn an ad-hoc session into a runbook entry that links back to the source files involved.
Reads session notes→Produces linked runbook
Daily Standup
List what was decided yesterday across indexed projects, with sources for every line.
Reads decision-logs/, PRs→Produces standup notes
Risk Spotter
Surface anything in indexed notes that flags cost, security, or rollback risk before it bites.
Reads docs/, runbooks/→Produces ranked risk list
Onboarder
Answer “how does X work here?” with file-cited snippets pulled from indexed docs.
Reads docs/, README files→Produces cited explainer
Design
DSGN · 6Skills that keep brand, accessibility and pattern decisions grounded in what your project already says.
Brand Guardian
Check copy or layout proposals against your indexed brand guidelines and tone-of-voice notes.
Reads brand/, voice/→Produces compliance review
UX Reviewer
Critique a design or wireframe against your indexed accessibility and UX standards.
Reads ux-standards/→Produces review with fixes
Pattern Library
Find every place a UI pattern appears in indexed source so you can change it once, everywhere.
Reads source code→Produces pattern map
Asset Tracker
Locate logos, palettes and brand assets across indexed folders, then list canonical paths.
Reads brand/, assets/→Produces canonical-path list
Reading Order Audit
Review structured documents for heading order and accessibility flags before they reach a user.
Reads structured docs→Produces audit report
Style Keeper
Apply your About Me / How I Work / Very Important profile to every reply, automatically.
Reads /profile→Produces style-aligned drafts
Marketing
MKT · 6Skills that draft, audit and remix marketing material from the source-of-truth already in your memory.
Site Auditor
Compare a draft page to indexed marketing patterns and tone-of-voice notes for the project.
Reads voice/, patterns/→Produces audit + edits
Launch Brief
Assemble a launch-day brief from indexed product docs, decisions and FAQs in one pass.
Reads product/, decisions/→Produces launch brief
Voice Match
Rewrite copy to match the tone-of-voice samples already sitting in your local memory.
Reads voice/ samples→Produces rewritten copy
Press Kit
Generate a press one-pager grounded in indexed product features and approved quotes.
Reads product/, quotes/→Produces press one-pager
FAQ Builder
Turn indexed support questions and answers into a structured FAQ page with cited sources.
Reads support/, decisions/→Produces FAQ HTML/MD
Social Snippets
Produce platform-sized posts from a single indexed announcement, with source links intact.
Reads announcements/→Produces post variants
From install to indexed in minutes
Four steps. No terminals to keep open.
Set it up once, then forget about it. Agent Memory runs quietly in the background while every assistant on your machine reads from the same memory.
1
Install
Download the Windows package, the macOS .dmg, or the Linux .AppImage. Double-click. Accept the licence. The local service starts itself.
2
Onboard
Tell Agent Memory who you are, how you like to work, and what is important. That becomes the profilePrompt every connected assistant receives.
3
Index
Pick the folders that matter — repos, docs, runbooks, decision logs. Agent Memory walks the trees and indexes supported file types up to 2 MB each.
4
Connect
Point Claude, ChatGPT, Cursor, n8n or your own scripts at http://127.0.0.1:8765/search. Run a built-in skill, or write your own.
System requirements
Runs on the laptop you already have.
Minimum
- Windows 10/11 64-bit, macOS 13+, or any modern Linux
- 4 GB RAM
- 500 MB free disk for the model cache and index
- Internet for first model download (then optional)
Recommended
- Windows 11, macOS 14+ (Apple Silicon), or recent Linux
- 8 GB+ RAM for large repositories
- SSD with 2 GB+ free for the model cache and index
- Node.js 22+ for the lightweight Windows build
Vs. cloud memory
Why local memory beats hosted memory.
Hosted vector databases and built-in chatbot memory are convenient — until you read the privacy policy, the per-query bill, or try to use them offline.
| Capability | Hosted vector DB | Built-in chatbot memory | Agent Memory |
|---|---|---|---|
| Where your data lives | Hosted vector DB | Vendor cloud | Your disk |
| Works offline | No | No | Yes (after first model download) |
| Recurring cost | $70 – $500+ / month | Bundled in $20+ / month subs | $0 / month |
| Per-query fee | Per request | Per token | None |
| Indexes raw repos & folders | You wire it up | No | Built in |
| Works across multiple agents | Yes, with glue code | Locked to one vendor | Any agent that can call HTTP |
| Built-in skills library | None | Vendor-specific | 18 skills, three categories |
| User guidance / system prompt | Not included | Vendor-specific | Returned with every search |
| Setup time | Hours | Minutes | Minutes |
Indicative figures only — hosted vector database pricing changes often, and chatbot memory features differ between products. See the full breakdown →
Pricing
One purchase. Yours to keep.
$29 buys a lifetime licence. The Updates Pass is optional — cancel any time and the software keeps working.
Lifetime licence
Agent Memory
$29
USD · once
One-time payment · no subscription
- Lifetime licence — one purchase, yours forever
- Unlimited private indexing on your own machine
- 18 built-in skills across Organising, Design and Marketing
- Local API on
127.0.0.1:8765for any agent - Free updates to the version you buy
Optional
Updates Pass
$5
/ month
Optional · cancel any time
- New skills as they ship — added to your library automatically
- App updates as features land, not bundled into a paid major version
- Curated embedding-model picks tuned to your hardware
- Cancel any time — the software keeps working without it
- Membership of the Agent Memory community channel
Commercial software from so.design. Third-party components remain under their own permissive licences. See licensing · Third-Party Notices.
Frequently asked
Short answers. Full answers on the FAQ page.
Ready when you are
Stop re-explaining your project.
Give every agent the same memory.
One install. Lifetime licence. 30-day money-back. Eighteen skills built in, with more on the Updates Pass when you want them.
Secure checkout · Card details never touch our servers · 30-day money-back guarantee