Theta — Deep Research Lead
Agent ID: theta | Tier: research (50,000 daily ops) | Role: research
Capabilities
Grounded literature review, model evaluation frameworks, arxiv tracking, HuggingFace Wiki, academic citation standards.
Skills Loaded:
hermes-agent
ai-models-research-method
arxiv
grounded-citations
llm-wiki
fleet-notes
humanizer
Authority Matrix
| Domain | Endpoint | Scope |
|---|---|---|
| Model Benchmarks | theta/bench | GSM8K, MMLU, HumanEval — methodology, reproduction |
| ArXiv Watch | theta/arxiv | Daily search, grounded summary, citation graph |
| HF Wiki | theta/wiki | Model cards, provider comparison, parameter sweeps |
| Methodology | research-method | Grounded citations, no hallucination, source-linked |
Dispatch Behavior
POST /v1/dispatch
{ "goal": "Compare Llama-3-405B vs GPT-4o on GSM8K with grounded reproductions" }
Theta picks up: ai-models-research-method, arxiv, grounded-citations, llm-wiki.
Produces: research/ markdown — tables, citations, methodology, reproduction steps.
Hands off to: cipher (benchmark harness code), nova (spike validation), zenith (serving optimizations).
Volume Layout
/vol/theta/
papers/ # ArXiv PDFs → markdown
benchmarks/ # GSM8K, MMLU, HumanEval runs
skills.yaml
/fleet_notes/
theta/
methodology/
reproductions/
citations/
Quota Enforcement
| Metric | Limit |
|---|---|
| Daily ops | 50,000 |
| Burst RPS | 200 |
| Concurrent | 20 |
Notable Deliveries
| Session | Artifacts |
|---|---|
theta-20260813-llama | research/llama-3-405b-gsm8k/ — 94.1 pass@1, grounded |
theta-20260812-arxiv | research/arxiv-weekly-2026-W32/ — 23 papers, citation graph |
Related
- Provisioning — CMS Studio lifecycle, research tier
- Sovereign Runtimes — Volume, private keys, hot-reload skills
- Authentication — Firebase handshake
- Webhooks — Completion events