# I gave Claude Code a memory and it used 26% fewer tokens

> Source: <https://omnimem.org/>
> Published: 2026-11-18 08:54:45+00:00

**Recommendation:** Persist past failed attempts and project context in an MCP memory server so agents skip known dead ends; the source measured 26% fewer tokens and 0 of 10 dead-end runs versus 8 of 10 without memory.
More in the playbook "Controlling LLM cost": <https://wpnews.pro/playbooks/llm-cost>

Stop living the same session twiceSelf-hosted memory for Claude Code and every MCP agent

OmniMem is an open-source MCP memory server that gives Claude Code, Codex, Cursor and any other MCP agent persistent memory across sessions, projects and machines. One app for your desktop or one binary on a server, everything on your own hardware and, yes, it's free.

Your AI agent doesn't know your project. It doesn't remember what failed last week. It has no idea you spent three hours last Tuesday figuring out why WebKitGTK quietly ignores a redirect from a custom URL scheme, only to find something that actually works.

So you explain everything again. It suggests the same broken library again. Same alarm, same song. It's Groundhog Day: you're Bill Murray, and your agent is Punxsutawney.

// without omnimem

claude>answer the settings form with a 303 redirect

... 45 minutes of debugging later ...

form submitted, window never moved (WebKitGTK, omnimem://)

You fixed this exact problem last Tuesday.

Claude doesn't know that.

// with omnimem

claude>answer the settings form with a 303 redirect

⚠ WARNING: previously abandoned approach

303 from omnimem://: WebKitGTK never follows it (effort 4/5)

→ answer with a page that replaces itself instead

Measured

Fewer tokens, smaller bills, less waiting

We gave Claude Code the same task twenty times on a fresh copy of the code: ten runs without OmniMem, ten with it. The task's TODO recommends a crate that fails at runtime, and only running the tests shows it. With OmniMem the earlier failure is already in memory, so the agent skips the dead end and gets on with the job.

Tokens

−26%

833k → 616k median tokens per run, fewer in 8 of 10 pairs.

without

OmniMem

Money

−26%

Median cost per run down by a quarter: $2.16 → $1.51 for all ten runs.

without

OmniMem

Time

−35%

Median wall time per run, because nobody spends a test run finding out the crate is broken.

without

OmniMem

Dead ends

0 of 10

Runs that walked into the known dead end, against 8 of 10 without. Red test runs went from 7 to 0.

without

OmniMem

OmniMem 7.0.12, Claude Haiku 4.5, one scenario and ten pairs, so treat the token saving as promising rather than proven (a sign test gives p = 0.05); the dead-end result doesn't depend on it. A dead end that takes longer to find would be worth more. Read the full report (PDF), which also covers contradictions, forgetting and project isolation.

See it in action

One recall() pulls context from everywhere

Personal preferences, ingested articles, past conversations, project context. It all gets ranked together and merged into one properly useful answer.

claude · recall

› what tools do we know about that are useful for writing markdown in?

- BookStack (MIT, PHP): wiki with markdown supportknowledge

- Wiki.js (AGPL-3.0, Node.js): wiki with markdown editor

- Docmost (AGPL-3.0, Docker): Confluence alternative

CMS with markdown support:

- Sveltia CMS and DecapCMS: both work with Forgejo/Gitea backendsproject

That's the lot, pulled from your preferences, your articles, a repo you trained it on, and your project's own stack, all ranked together.

› nope that's great, you reminded me of exactly what I use

● Glad it worked well. OmniMem connected your preferences and past writing into one useful answer.

A real recall() about markdown editors. Every tagged line traces back to a different memory source (they're the four on the right).

[episodic]

Episodic memory

Picked up from your conversations: decisions, bugs, what you tried, and the things you mention in passing.

[knowledge]

Knowledge base

RSS articles, auto-summarised and embedded. They surface when relevant to what you're asking.

[preference]

Preferences

How you like to work, stated once and applied from then on: the editor you reach for, the licence you prefer, the way you want things written.

[project]

Project context

Extra context it knows matters to you, like your preferred tools and platforms.

What makes it different

Not just a key-value store with an MCP wrapper

OmniMem models how memory actually works. Things fade over time, they sometimes contradict each other, and the hard-won stuff earns its place.

The graveyard

Every dead end, kept on purpose

What you tried, why it failed, and how much time it burned. Your agent checks the graveyard before it reaches for a library or a pattern, so the same rabbit hole only costs you once.

A warning comes back with the way out attached: what was abandoned, and what worked instead.

warn_if_abandoned("global limiter") ✗ abandoned· effort 4 · 6 tries → use one limiter per shard instead

recall()

One call. Ranked by similarity, lifecycle, age and what it cost you to learn.

48

MCP tools

Remember, recall, audit, compile. One server, every client that speaks MCP.

1.8×

Battle-hardened

Four attempts and a weird workaround? That is gold. The harder it was, the higher it surfaces.

Four namespaces

Searched together, never muddled

Decisions and bugs, stack and state, articles and extracted facts, and how you like to work. Compiled skills sit alongside them.

episodicprojectknowledgepreference

One-call briefing

No three-step warm-up

One briefing() hands over project context, experience stats, stale memories, new articles and contradictions.

If a new memory disagrees with one you already have, OmniMem catches it on write, and Claude can dig deeper on demand.

Deduplication

Semantic, not string

Too close to something stored? You get a heads-up, not a duplicate. find_duplicates cleans up in bulk.

Settings window

A window for all of it

Browse, search and manage memories, feeds, projects, skills and backups from the tray icon, without touching the command line or opening a port.

Telemetry

Know what earns its keep

Recall counts and last-accessed per memory, what's gone cold, and how much of your agent's context OmniMem takes before a single tool is called.

Auto-maintenance

Tidies itself

Every few briefings it scans for duplicates and checks for conflicts in the background. No babysitting.

Lifecycle

Memory is not binary

Most systems either remember or delete. OmniMem has a proper lifecycle. When you say "forget about X" you usually mean stop bringing it up, not wipe it from existence.

ACTIVE

1.0× weight

→

DEPRIORITISED

0.2× weight

→

ARCHIVED

0.0× weight

→

DELETED

gone

Deprioritised memories aren't gone for good. Attach reinstate hints, and if a future query matches, the memory comes back with a note explaining why it was pushed down. You can mute entire topics across all sessions.

Five factors decide what comes back. Semantic similarity on its own isn't enough, so lifecycle state, age, how hard something was to figure out, and whether it happened near a date you asked about all play a part.

Effort

Meaning

Weight

1

Worked first time

1.0×

2

Minor friction

1.1×

3

Multiple iterations

1.25×

4

Significant struggle

1.5×

5

Battle-hardened

1.8×

Skills

Skills that build (and evolve) themselves

Memories tell an agent what happened. A skill tells it how you work. OmniMem compiles skills from the work you actually did: the breakthroughs, the gotchas that keep coming back, and the graveyard of dead ends. As new experience lands, the skill evolves with it.

A PATTERN EARNS A RULE

One-off episodes don't compile. A lesson has to recur before it becomes policy, unless you bless it to jump the queue.

NOTHING WRITES SILENTLY

Every compile proposes a diff you can review. Rewrites of existing rules get flagged loudly, and nothing commits until you say so.

DERIVED, NEVER HAND-EDITED

Memories are the source of truth and the skill is just build output. Want different guidance? Update the memories and recompile.

SUGGESTED, NEVER AUTO-LOADED

The briefing recommends relevant skills for the session. You and the agent decide what loads.

compile_skill("python") distils a domain's accumulated experience into a loadable SKILL.md: do this, watch out for that, and never try X again because it cost you an afternoon. Load it at the start of some Python work and the agent works your way from the very first prompt.

// the flywheel

work → record_experience() → compile_skill() → SKILL.md → work

Every skill tells the agent to keep recording lessons and dead ends whilst it works. A richer pool compiles a better skill next time, and the knowledge watch flags fresh articles that contradict a compiled rule, so your skills stay current.

✓ SKILL.md committed · every rule cites its memories

How it compares

Most memory servers just store and retrieve

Claude Code's built-in memory (CLAUDE.md, plus the newer auto-memory that writes markdown notes for itself) is still flat, machine-local files with keyword recall, as persistent memory for Claude Code explains in more detail. No semantic search, no lifecycle, no contradiction handling. OmniMem goes quite a bit further. mem0 is the other name you'll hear. It's a good product aimed at a different job, a memory layer for products serving many users with a managed cloud behind it, and the table includes the things it does that OmniMem doesn't.

A green box with a tick matches or beats OmniMem on that row, and ~ marks a partial one. A black box in the OmniMem column is a straight no where another system has it. Compared in September 2026, against each project's own documentation.

Local embeddings, one SQLite file, one process. The MCP server, the RSS scheduler and the enrichment worker share one memory engine, and the desktop app's settings window talks to it directly, so there's no dashboard port for anyone else to find.

RSS scheduler · enrichment worker ONNX embeddings, no GPU Claude Haiku, if you give it a key

// recall pipeline

→ query

→ abandoned fast-path (keyword scan)

→ embed query

→ exact vector search, 20+ per namespace

→ filter archived + deleted

→ filter out-of-project + suppressed topics

→ apply surface_score

→ apply recency (0.9 at 90 days)

→ apply experience_weight

→ apply date_boost

→ check reinstate eligibility

→ surface contradiction warnings

→ merge namespaces, facts fold into their sources

→ drop anything under the relevance floor

→ re-rank, return top_k

→ log event + increment counters

Get started

Up and running in two minutes

Pick the app for your computer, or the server if your agents live elsewhere. Nothing else to install: no Docker, no database, no Python. The embedding model fetches itself on the first start.

01

Install it

Download the app: drag it to Applications, run the MSI, or open the Flatpak. It lives in your menu bar or system tray from then on. For a server, install the .deb or .rpm from the release page, or run the Docker image.

a Linux servercopy

```
sudo apt install ./omnimem_<version>_amd64.deb
# on Fedora: sudo dnf install ./omnimem-<version>-1.x86_64.rpm
curl http://127.0.0.1:8765/healthz   # {"status": "ok"}
# settings live in /etc/omnimem/omnimem.env
```

02

Connect your agent

In the app, the tray menu's Add to Claude Code and Add to Claude Desktop do it for you, and never add it twice. Copy MCP URL covers every other agent. By hand, for Claude Code:

bashcopy

```
claude mcp add --transport http --scope user omnimem http://127.0.0.1:8765/mcp
# on a server with a token, add: --header "Authorization: Bearer <token>"
```

03

Add an Anthropic key optional

OmniMem works fine without one. With a key, Claude Haiku pulls the facts out of what your agent saves, rephrases recall queries so they find more, double-checks contradictions and summarises your RSS feeds. Paste it in the settings window's Configuration page, under Claude, and it takes effect at the next start. It's kept in your OS keychain. On a server, set ANTHROPIC_API_KEY in omnimem.env. Embeddings still run on your machine either way.

Coming from 6.x? Stop it first, since both use port 8765. Take a backup on 6.x with dump_to_file(), then restore it on the settings window's Backups page or run omnimem import memory_backup.json. Every memory is embedded again from its text, so recall and skills behave as they did.

Pro tip

Agents mostly don't ask their memory anything, so for Claude Code there are hooks: omnimem hook puts the project's briefing in every session's opening context and stops a tool call that repeats a known dead end before it runs.
