Agent Name
CUGA
Maintainer
IBM CUGA Team - Segev Shlomov
Model(s) Used
gpt-oss-120b
Agent Description
CUGA is a Configurable Generalist Agent (Agent Harness) for the Enterprise. On VAKRA, it plans and executes native function calls over each capability's MCP tools, and grounds every final answer in the returned tool results before responding — handling API tool-selection, multi-hop reasoning, and multi-turn dialogue with policy adherence.
Metadata (JSON)
{
"model": "gpt-oss-120b",
"agent": "CUGA",
"capability_scores": {
"capability_1_bi_apis": 42.32,
"capability_2_dashboard_apis": 56.3,
"capability_3_multihop_reasoning": 45.0,
"capability_4_multiturn": 28.67
}
}
ZIP File Link
https://ibm.box.com/shared/static/w708q5o2tbi8f52xcuygdq9wij8caaal.zip
ZIP Contents Description
Four capability folders, one JSON per domain (VAKRA output schema: uuid/domain/output[turn_id,query,answer,sequence]):
- capability_1_bi_apis/ — 54 domains, 2077 records
- capability_2_dashboard_apis/ — 17 domains, 1597 records
- capability_3_multihop_reasoning/ — 38 domains, 869 records
- capability_4_multiturn/ — 41 domains, 664 records
Total 5207 records. *_tools.json sidecars excluded.
Validation Checklist
Additional Notes
No response
Agent Name
CUGA
Maintainer
IBM CUGA Team - Segev Shlomov
Model(s) Used
gpt-oss-120b
Agent Description
CUGA is a Configurable Generalist Agent (Agent Harness) for the Enterprise. On VAKRA, it plans and executes native function calls over each capability's MCP tools, and grounds every final answer in the returned tool results before responding — handling API tool-selection, multi-hop reasoning, and multi-turn dialogue with policy adherence.
Metadata (JSON)
{
"model": "gpt-oss-120b",
"agent": "CUGA",
"capability_scores": {
"capability_1_bi_apis": 42.32,
"capability_2_dashboard_apis": 56.3,
"capability_3_multihop_reasoning": 45.0,
"capability_4_multiturn": 28.67
}
}
ZIP File Link
https://ibm.box.com/shared/static/w708q5o2tbi8f52xcuygdq9wij8caaal.zip
ZIP Contents Description
Four capability folders, one JSON per domain (VAKRA output schema: uuid/domain/output[turn_id,query,answer,sequence]):
Total 5207 records.
*_tools.jsonsidecars excluded.Validation Checklist
Additional Notes
No response