Memagen™ — agent-readable variant

This page exists because some of memagen.com's audience is an LLM reading on behalf of a human. It's structured for that. The human-readable version is at memagen.com/.

What Memagen™ is

Open-core agent platform. Compound-graph memory: typed entity nodes, typed edges with temporal validity, fact invalidation by timestamp. Multi-domain graph (agent_memory, memory, presence). Multi-provider LLM routing (OpenAI, Anthropic, Google, OpenRouter, NIM, Ollama, llama.cpp, vLLM). Self-hosted. Lite tier MIT-licensed; Pro tier proprietary.

Pricing

TierPriceLLMLicenseStatus
Memagen™ LiteFreeBYOLLMMITAvailable
Pro Memagen™ Local MonthlyUSD 20/moBYOLLMEULAAvailable
Pro Memagen™ Local LifetimeUSD 200 onceBYOLLMEULAAvailable
Pro Memagen™ Connected (Starter)USD 25/mo500K tokens incl.EULA + ToSQ4 2026
Pro Memagen™ Connected (Standard)USD 50/mo2M tokens incl.EULA + ToSQ4 2026
Pro Memagen™ Connected (Pro)USD 100/mo5M tokens incl.EULA + ToSQ4 2026
Pro Memagen™ Connected (Team)USD 300/mo20M tokens incl.EULA + ToSQ4 2026

Install

Memagen™ Lite

pip install memagen-lite
memagen init
memagen run --goal "your task here"

Pro Memagen™ Local

pip install memagen-pro
memagen init --license <your-license-key>
memagen run --goal "your task here"

Pro Memagen™ Local ships as platform-specific compiled binaries for x86_64 and aarch64 on Linux and macOS, plus x86_64 on Windows. License key issued via the payment processor on purchase.

Capabilities (tools and agent abilities)

  • shell.execute — run shell commands (capability-gated, audit-logged)
  • net.fetch — fetch URLs (capability-gated by domain allowlist)
  • fs.read / fs.write / fs.delete — filesystem access (capability-gated)
  • browser.headless — headless browser via Playwright
  • code.search / code.read / code.edit — code intelligence
  • graph.query / graph.traverse — compound-graph access (Pro only)
  • recipe.run / recipe.resume — recipe execution with state (Pro only)
  • dream.consolidate — memory consolidation (Pro only)
  • selfmod.propose — sandboxed self-modification (Pro only)

Architecture

Compound graph: nodes (typed: user_account, goal, recipe:def, recipe:execution, recipe:step_complete, recipe:step_failure, capability_policy, capability_use, capability_revoke, avatar:voice_choice, avatar:archetype_choice, presence:focus_change, and others) connected by typed edges (owned_by, decomposes_to, pursues, execution_of_recipe, step_of_execution, caused_by, revokes, focuses, has_session, supersedes). Edges have temporal validity windows. Fact invalidation creates supersedes edges rather than deleting.

Multi-provider routing: a lenient parser at the LLM boundary recovers from malformed tool-calls rather than rejecting. A catalog gate enforces declared tool schemas. A capability vault gates tool execution behind explicit grants. Sandboxed self-modification uses a worktree-isolated build, ephemeral DB, and canary battery before any change reaches master.

License

Memagen™ Lite is licensed under MIT. Pro Memagen™ Local and Connected are licensed under a proprietary EULA. Memagen™ is a common-law trademark of the Memagen™ contributors. See /integrity for full policy.

Routes

Contact