They built TRACER, which indexes observed tool-call sequences in a trie structure (full-prefix conditioning).
History-aware tool retrieval cuts LLM agent token costs by 80 percent while boosting accuracy
TRACER indexes tool-call sequences in a trie and learns to blend usage memory with semantic similarity, outperforming full-registry selection across most model-environment pairs.
Big Tech
Tanish Kher · Sarvesh DK · Abedelkadir Asi · Sparsha Chennarayapatna Sundaresh · Dezhi Hong
Microsoft
Research Digest··3 min read
Thread:Agent Harness Optimization
The authors propose TRACER, a retrieval front-end that prunes tool registries to a small candidate set by exploiting the sequential nature of tool calls in multi-turn agent tasks.
Why this paper
From Microsoft · Part of Agent Harness Optimization, now 143 papers
In one line
TRACER improves tool-selection accuracy and reduces token usage by using a trie of past call sequences to retrieve tools.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§