◇
Models need cognitive infrastructure
Better agents need more than a model. They need memory and state around the model.
Memory OS for coding agents
Dhee gives coding agents memory that actually sticks.
Portable memory, context routing, and shared state so agents stop replaying bloated history, resume real work, and collaborate without an orchestrator agent.
Dhee memory runtime
00.00[system] Coding agent attached. Session identity verified.
00.11[dhee] Restoring prior tasks, artifacts, rules, and durable memory.
00.24[router] Relevant context loaded. Wasteful history held back.
00.37[agents] Shared handoff open. No orchestrator agent required.
00.48success Work resumed with portable state and live continuity...
Less junk in context
Load what matters now, not the full transcript every time.
Gets sharper with use
Retrieval improves as agents expand, reuse, and hand off work.
No boss agent needed
Shared memory lets multiple agents coordinate directly.
Philosophy
Why Dhee exists
Session memory, giant transcripts, and brittle orchestration are not enough for real software work.
◈
Memory has to be owned
If memory lives inside one provider session, it gets messy fast. Dhee keeps it portable.
◉
Agents should evolve through use
Dhee does not just store memory. It learns how to serve it better over time.
How Dhee works
Three layers turn a coding agent into a personalized, self-evolving collaborator.
Dhee adds memory, routing, and shared state around the model.
Memory substrate
Durable state outside the modelTasks, artifacts, and decisions survive past a single run.
Context router
The right slice at the right timeDhee decides what the active agent should see now and what should stay compressed.
Shared agent fabric
Collaboration without an orchestratorAgents can hand off work through shared memory instead of a separate orchestrator.
Why now
Developers are still burning tokens on wasteful context.
Long tasks and multi-agent work still break because memory and context handling are weak.
Session memory is not ownership
Most agent memory disappears the moment the session or provider changes.
Raw history is not useful context
Full transcript replay is noisy, expensive, and usually unnecessary.
Orchestrator agents become bottlenecks
A central planner adds overhead. Shared state scales better.
Our response
Own the layer between the model and the work.
Dhee is the memory layer you own. It keeps agents from repeatedly loading bloated context, learns from what actually gets used, and lets multiple agents coordinate through shared memory instead of orchestration overhead.
Portability
Built so the memory survives model churn
Memory stays stable even when providers and shells keep changing.
Efficiency
Efficiency without gimmicks
Better routing naturally means less wasted context.
Evolution
Use should sharpen the system
Usage teaches Dhee what to retrieve next time.
Who it serves
Built for developers doing real software work.
Use it when continuity, context quality, and handoffs matter.
Coding agents
Agents that need continuity
For coding agents that need to resume work without rebuilding context.
Project facts, decisions, and artifacts stay recoverable
Better resumes after interruptions or provider switches
Long tasks stay coherent without giant transcripts
Agent platforms
Teams building runtimes
For teams that want memory and context routing without a new framework.
Works through MCP, Python, CLI, and native integrations
Lets you keep your existing agent surface
Portable cognition layer across toolchains
Multi-agent teams
Workflows that outgrow central planning
For multi-agent workflows that need shared memory instead of a boss agent.
Any agent can resume from shared context
Artifacts and digests become coordination primitives
Collaboration scales without one controller agent
Next step