Built for builders
Poneglyph is designed for teams building with agents, where project knowledge and prior decisions need to survive beyond a single session.
Better memory for agents
Private, fast, absolute memory for your AI agents.
Poneglyph gives builders a memory layer for existing agents that stays local, recalls fast, and scales beyond markdown files and ad hoc note stores.
Works out of the box with OpenClaw, and is designed to become the memory layer for the rest of your agent stack too.
Poneglyph Agent memory
Project knowledge
Runtime is Rust. UI is Astro. Memory is shared between OpenClaw, Codex, and local tools.
Linked context
Packages, systems, concepts, and prior decisions stay connected instead of disappearing into notes.
npx -g i poneglyph
# connect OpenClaw or another agent
# write facts
# recall project knowledge laterPrivate by default
Memory stays local to your machine, which keeps recall fast and keeps sensitive context out of a hosted black box.
Structured to scale
Facts, entities, and relationships beat markdown folders once your projects, systems, packages, and people become deeply connected.
Quickstart
From zero to agent memory in a few commands.
Open source. Local-first. Start with OpenClaw or another agent and begin building project memory immediately. No hosted account required.
npx -g i poneglyph
# connect OpenClaw or another agent
# write facts
# recall project knowledge laterAbsolute memory
Your agents stop forgetting the shape of your projects.
Poneglyph gives agents durable memory across sessions, projects, and time. Less repetition, less context loss, less restating architecture by hand.
Existing agent stack
Works with the tools you already use.
Start with OpenClaw out of the box. Extend the same memory layer to Codex, Claude-based agents, and other MCP clients.
Project knowledge
A memory model built for systems, packages, and decisions.
Poneglyph models concepts, systems, packages, people, and project knowledge directly, so builders can recover how a project fits together.
Why not markdown
Markdown memory is elegant until it stops scaling.
Files are simple, but project memory eventually becomes deeply connected. Systems refer to packages, packages refer to concepts, decisions refer to tradeoffs, and agents need more than a folder tree to recover that shape reliably.
Why not SQL
SQL stores records well. Memory is a different problem.
Application tables are optimized around current state. Agent memory needs durable history, provenance, relationships, and replayable derived views. That is why Poneglyph uses an immutable graph model under the hood.
What changes
Without Poneglyph. With Poneglyph.
Without
Your agent forgets project decisions, package boundaries, and prior work between sessions.
With Poneglyph
Poneglyph gives it durable recall across time, so project context compounds instead of resetting.
Without
You keep restating the same architecture, constraints, and preferences every time an agent wakes up.
With Poneglyph
Poneglyph turns repeated explanation into accumulated memory your agent can recover later.
Without
Team knowledge lives in notes, logs, issues, and chat fragments that never form one coherent memory.
With Poneglyph
Poneglyph links project knowledge into a graph your agents can query instead of guess at.
Under the hood
Why the model is different.
Poneglyph handles long-term memory with facts, entities, and projections so the system can stay durable, inspectable, and replayable as it grows.
- Atomic fact writes with ordered transaction IDs
- Strongly consistent reads over facts after commit
- Asynchronous consolidation into emergent entities
- Projection workers for search and other read models
- Local service surfaces for agents and operator tooling
Get started
Install Poneglyph and give your agents better memory.
npx -g i poneglyph