Tencent Open-sources TencentDB Agent Memory, Reducing Token Consumption by Up to 61% in Long Task Scenarios

Tencent Cloud recently open-sourced TencentDB Agent Memory, which provides short-term memory compression and long-term personalized memory capabilities for Agent long-task scenarios. The long-term memory feature was launched last month for free, and this open-source release focuses on short-term memory compression technology.

Using the "Context Offloading + Mermaid Task Canvas" solution, the technology stores complete information in external systems while retaining key states and execution paths as structured task graphs. This keeps Agents lightweight in long tasks and allows layered traceability and recovery of original information. Experiments in multi-task continuous sessions show that this solution can reduce token consumption by up to 61% and improve task success rates in long-task scenarios.