Registration opening soon · Coming to  macOS  ·  Windows  ·  Linux
The memory layer for your AI

Your AI has
amnesia.
Anvil is the fix.

Every coding assistant forgets everything the moment you close it. Anvil is the private, encrypted brain that sits underneath — it remembers your projects, learns your style, catches its own mistakes, and hands all of it back to your AI on every prompt.

100% local & encrypted
200+ features · 80+ mentor skills
plugs into Claude Code & Desktop
works fully offline
See it think — live

This is Anvil's mind, actually running.

Not a mockup — a real 3D map of one Anvil brain, every memory, skill and project a star. Caught mid-thought: the Crucible's Embers debating a question around the core while housekeeping quietly works the local files. In the app you drag, zoom and fly through your own.

◆ live capture — the Crucible mid-debate
Forged on your side
Local-first

Your memory, projects and keys live on your machine, encrypted — nothing sent to a cloud you don't control. It's yours, on your desk, in a locked drawer.

Degrade, never brick

Lose a provider, a licence or a connection and you drop a rung — you're never stranded. And it's honest when something's off, rather than pretending everything's fine.

Free that helps

Anvil Free isn't a crippled trial — the whole memory brain is yours, forever. Mentors, not ghostwriters; guardrails included.

Why it exists

A chat window is not a memory.

The tools are incredible — and goldfish-brained. Close the tab and the context is gone: your architecture, the decisions you made, the bug you already fixed twice. So you re-explain yourself, forever. Anvil ends the loop.

Without Anvil

Re-paste the same context every session. The AI repeats mistakes it "learned" yesterday. Your knowledge lives in a cloud you don't control, or in scrollback you'll never find again.

With Anvil

A brain that persists. It recalls the right memory automatically, refuses to repeat a logged mistake, keeps secrets out of the model entirely, and shows you exactly what it knows — as a living 3D galaxy.

What's inside

Seven things it does that nothing else does together.

Built shipping today In Dev being built now Coming Soon on the roadmap
01 · Memory

A brain that actually remembers

Not a scratchpad — a structured, searchable long-term memory that gets denser and smarter over time.

  • Auto-recall — the right memory is injected into every prompt, no asking. Built
  • Knowledge graph — facts, entities & relationships, not just text blobs. Built
  • Contradiction detection — flags when a new fact clashes with an old one. Built
  • Time-travel & provenance — see what it knew, and when, and from where. Built
  • Right-to-forget + TTL — memories you can expire or delete for good. Built
  • Endless-session distillation — supports long-running chats compacted many times: a session you never close, grown to tens of thousands of turns, is folded into memory in full, never truncated. In Dev
02 · Privacy

Private by design, not by promise

It all lives on your machine, encrypted. Anvil endeavours to keep your secrets out of what's sent to the AI — a best-effort scrub, not a DLP guarantee.

  • Encrypted local store — AES-256, on your disk, yours. Built
  • Secrets vault — keys sealed away; kept out of the model's context on a best-effort basis. Built
  • Auto-redaction — scrubs API keys & tokens before anything is sent. Built
  • IP ringfence — your memories & skills can't be exfiltrated by a prompt. Built
  • Fingerprint/Face unlock — Touch ID & Windows Hello for the vault. In Dev
03 · Learning

It levels up from its own work

Anvil turns lived experience into reusable skills and hard rules — including the mistakes.

  • Skills-as-substrate — a growing library the AI reaches for by intent. Built
  • Rules-from-mistakes — a repeated error becomes a guardrail it can't cross. Built
  • Learn-from-memory — distills your corpus into skills & rules automatically. Built
  • Hot-load skills — add new skills live, no restart. Built
  • Self-optimizing personas (GEPA) — roles that measurably improve over time. Built
  • Crucible — a cast of Embers (characters you direct, fused with real skills) argue an idea & resolve it to a sharper answer. Built
04 · See it think

Watch the brain, live

Most AI is a black box. Anvil renders its mind as something you can actually look at.

  • 3D knowledge galaxy — every memory & skill as a star you can fly through. Built
  • Live flow-graph — see providers, roles & skills fire in real time. Built
  • Click-to-toggle — flip providers, roles & skills on the fly. Built
  • Cost & token meter — exact spend per task, no surprises. Built
05 · Run the work

Your projects, not just your prompts

A whole project brain — plans, tasks and architecture — wired into the same memory.

  • Tasks & kanban — first-class task board with priorities & projects. Built
  • Blueprint — a live map of what your system is actually made of. Built
  • Roadmap tracks — phases, milestones & what's gating what. Built
  • Reading list — capture a URL to research in one tap. Built
  • Full PM suite — dashboards, dependencies, ownership & audit. Coming Soon
06 · Run anywhere

Cloud when you want, offline when you don't

Bring your own models. Anvil routes across providers and falls back gracefully — even to your laptop.

  • Multi-provider routing — with automatic failover & cooldowns. Built
  • Fully offline tier — local models (MLX & Ollama), no internet needed. Built
  • Rate-limit & cooldown visibility — never get blindsided by a cap. Built
  • Native Windows — runs on named pipes, no WSL required. Built
  • Local voice (TTS) & on-device images. Coming Soon
07 · Trust & teams

Built to be trusted with real work

Grounded answers, honest limits, and — for teams — the governance a company needs.

  • Grounding trust — it won't dress up a guess as a verified fact. Built
  • Capped autonomy — it can't self-approve risky changes; you gate them. Built
  • Shared team memory & encrypted sync — a central brain for the crew. Coming Soon
  • Users, roles & access control (IAM) — Cedar-powered, read/change/admin. In Dev
  • Forensic audit log — tamper-proof, encrypted, every action accountable. Coming Soon
08 · Security

It hunts for holes — including its own

A full offensive-and-defensive security range built in, plus an audit Anvil runs against itself.

  • Proving Ground — a gated recon → hunt → prove → report engagement against a target you're authorised to test. Built
  • Six hunt playbooks — access control, auth & tokens, web/client, injection, LLM app-sec, SSRF & infra. Built
  • Triage gate — a 7-question test + CVSS rubric kills false positives before they're ever reported. Built
  • Self-security assessment — Anvil audits its own MCP, vault & token custody, and has closed real findings. Built
  • Threat modeling — STRIDE, trust boundaries & attack trees, risk-ranked. Built
The complete picture

See the full feature set →

Every capability and its current state — Built, In Dev, or Coming Soon. 200+ features, honestly labelled.

How it stacks up

Anvil vs. everything else you've tried.

The AI assistants are great at answering. The memory add-ons are cloud lock-boxes. Anvil is the only one that's a private, structured brain and wired straight into your agent.

Capability Anvil Raw AI assistant
Claude Code · Cursor · Copilot
Memory add-ons
ChatGPT memory · Mem0 · Letta
Persistent cross-session memory deep & structured forgets on close shallow, per-account
Local, encrypted memory store on your disk cloud cloud
Knowledge graph (facts + relationships)◑ some
Every memory is cited & traceable grounded + provenance no memory◑ opaque
Catches its own contradictions
Learns skills & rules from its own work
Multi-persona reasoning debate the Crucible
Independent verifier on its own output
Built-in security range (red-team + self-audit) proving ground
Best-effort secret scrubbing vault + scrub
You own your data — expire or delete anything right-to-forget + TTL◑ their servers
Visual 3D map of what it knows the galaxy
Project management built in tasks · blueprint · roadmap
Works fully offline on local models◑ some
No vendor lock-in — routes across providers + failover single vendor
Keeps working if your subscription lapses offline licence
Team brain + roles + audit shipping in tiers◑ enterprise plans
Where we sit in the scene

Good company, different lane.

We're genuinely fans of a lot of these tools — some we even run on top of. Here's how we see the landscape, no shade intended.

Hermes Nous Research · hermes-agent

Their strength: reach and ecosystem — messaging channels, a huge model portal, agent training data, delivery everywhere an agent can show up.

Our angle: we go the other direction — deep on a private, encrypted, rigorously-structured memory brain. Anvil even leads on the trust bits (vault, signed context, contradiction detection, right-to-forget). Different lanes; plenty of room for both.

AI IDEs Cursor · Windsurf · Copilot

Their strength: unreal at writing code in the moment — fast, in-editor, great autocomplete and edits.

Our angle: they don't keep a private, structured memory of your work between sessions. That's the layer Anvil quietly adds underneath.

Claude Code & Desktop we plug right in

Their strength: the assistant you already love using every day.

Our angle: not a rival — Anvil connects straight in as the brain, so the tools you have suddenly remember.

Memory add-ons Mem0 · Letta · ChatGPT/Claude memory

Their strength: same good instinct as us — give the AI a memory so it stops starting from zero.

Our angle: we keep it local, encrypted and structured into a graph, not a cloud key-value box you don't own.

Local AI studios LM Studio · Ollama

Their strength: brilliant at running models right on your machine.

Our angle: Anvil sits a layer up — the memory, learning, projects and governance around the model. (We happily run on top of Ollama, in fact.)

Notes + AI Obsidian · Notion

Their strength: lovely homes for notes you file and organise yourself.

Our angle: Anvil reads them in and feeds them to your agent automatically — the curating happens for you.

The stuff we don't even shout about

Stealth features already shipped.

Small things that quietly make Anvil feel like it was built by people who've actually shipped software.

Forced-recall hook

Grounding is injected on every prompt — the AI can't "forget" to check its memory.

Import your past

Pull in Obsidian, OneNote & your whole chat history on first run.

Interactive tool loop

A reasoning turn can go fetch its own data mid-thought.

Authenticated browsing

Reaches logged-in pages (Playwright) when research needs it.

Circuit breaker

After N failed attempts it stops and escalates instead of looping forever.

Memory lint

Scans its own store for health — dupes, dangling links, stale facts.

Redaction preview

See exactly what gets scrubbed before anything leaves your machine.

Content-free ops journal

Diagnostics that never leak what you were actually working on.

Backup & restore

Export/restore your whole brain — a real data-loss guard.

Plans & features

The full brain is free. Pro sharpens it. Enterprise governs it.

One download, three plans — the same app, unlocked by your licence. Your memory is never gated: the local encrypted brain is yours in full on every plan, forever. And it degrades gracefully — a lapsed licence or a downed provider drops you a rung, it never leaves you stranded.

Free
The whole memory brain, on your machine, forever — not a crippled trial.
  • Full local encrypted memory + knowledge graph — unlimited
  • Auto-recall injected into every prompt
  • Contradiction detection + right-to-forget (TTL)
  • The free built-in success-stack skills
  • 3D knowledge galaxy + live flow-graph
  • Tasks with priorities
  • One AI provider — bring your own (cloud, or fully offline via MLX & Ollama)
  • Secrets vault · best-effort redaction · IP ringfence
  • Grounded, cited answers · backup & restore
Pro
Everything in Free, plus the tools for people who live in this all day.
  • The Crucible — convene a council of Embers on the hard calls
  • Custom & Pro skills — author your own, the full skill set
  • Multi-provider routing + automatic failover & cooldowns
  • Blueprint — Map · Diagram · Tasks · Roadmap
  • On-device image generation
  • Project planning
  • External trackers — Jira · Linear · GitHub
  • Self-optimizing personas (GEPA) Built
Enterprise
Everything in Pro, plus a shared, governed brain for the whole team.
  • Shared / central team brain
  • Shared kanban board
  • Users, roles & access control (RBAC)
  • Forensic audit log
  • SSO / LDAP & central admin
  • End-to-end encrypted sync
◇ Enterprise is on the roadmap — in design, not yet shipped.
Free & Pro features shown here ship today (except items tagged In Dev). Enterprise capabilities are planned, not yet available. Pricing announced at launch.
What's next

Coming soon.

On the roadmap, not yet in build — the direction we're forging toward.

Forensic audit — encrypted, tamper-proof, AI-action accountability
Full IAM — SSO, LDAP & central admin
Host-anywhere team server — local, VPS, your call
Crucible multi-model — assign Embers to different providers
Local voice (TTS)
On-device image generation
Jira & external trackers
PM dashboards & burn-up
Syntax highlighting in chat