Capture the signal.
Collect the decisions, outcomes, and preferences worth carrying forward—not every token of every conversation.
EVENTS → CONTEXTTHE MEMORY LAYER FOR AI AGENTS
Give your agents
memory that lasts.
Useful context shouldn’t disappear when a session ends. OpenMem is building a memory layer that helps agents carry forward what they learn—and bring back only what the next task needs.
01 CONTEXT THAT CARRIES
02 RECALL WITH INTENT
03 SOURCES YOU CAN TRACE
MEMORY / RECALL
Pick up where you left off.
I found the migration decision, the team’s rollout convention, and one remaining check. Here’s the short version—with the source trail attached.
RECALLED CONTEXT 3 sample memories · ranked for this task
FIG 01 An agent asks. A compact, traceable context comes back. INTERACTIVE SAMPLE ↗
01 A SIMPLE MEMORY LOOP
Add. Learn. Retrieve.
RAG helps an agent find useful information in the world. Memory helps it carry forward what happened in its own work. They solve different problems—and work better together.
Keep what changes the next step.
Group related events, revisit what’s still true, and connect each useful memory back to where it came from.
CONTEXT → MEMORYBring back just enough.
Find context for the current task, rank it by usefulness, and fit it to a budget the agent can control.
MEMORY → NEXT ACTION01 RETRIEVAL-AUGMENTED GENERATION
Find what the world knows.
Search external sources—documents, knowledge bases, and the web—for information relevant to this prompt.
EXTERNAL KNOWLEDGE ↗02 AGENT MEMORY
Carry forward what work taught.
Bring relevant decisions, outcomes, and preferences from an agent’s own prior interactions into the next task.
EXPERIENCE OVER TIME ↗OpenMem is in development, exploring how these two kinds of context can work together.
02 MADE FOR LONG-RUNNING WORK
A short memory is a short leash.
When an agent returns to a task, it should not have to rediscover the same decision. OpenMem is being designed for systems that need continuity across runs, tools, and time.
Tell us what you’re buildingCoding agents
Carry project conventions, prior debugging steps, and decisions between work sessions.
Research agents
Keep track of what’s been checked, what changed, and which questions remain open.
Support & operations
Give an agent useful history across interactions without replaying an entire thread.
03 DESIGNED WITH GUARDRAILS
Trust needs a source.
Useful memory should be inspectable—not mysterious. The product direction is built around clear boundaries and explicit control.
- 01Trace every memorySee the event or source behind a recalled detail.↗
- 02Scope context deliberatelyKeep agent, session, and workspace context distinct.↗
- 03Make lifecycle a first-class concernInspect, revise, or remove what an agent has retained.↗
These are design goals for the product in development—not claims about a currently available service.
OPENMEM / EARLY ACCESS
Help shape agent memory.
We’re speaking with teams building agents that need to remember more than a single session. Tell us what you’re working on; we’ll share updates as the preview takes shape.