PACK 01
Portfolio
Allocation, exposure, concentration, risk constraints, rebalance generation, and structured financial outputs.
1B–7B target8K contexttool-aware
PANDOS MODEL PACKS
Small models.
Actual jobs.
We are building specialist model families around specific agent workloads: training and fine-tuning compact models, distilling stronger intelligence into smaller runtimes, evaluating them against job-specific benchmarks, and distributing local and managed builds through PandOS.
Domain datasetsSFT / distillationTask evalsQuantizationLocal buildsCloud builds
PACK 02
Research
Retrieval, extraction, document reasoning, source comparison, synthesis, and citation-oriented output.
local-firstRAG readyfile tools
PACK 03
Planning
Goal decomposition, prioritization, dependency resolution, scheduling, and structured execution plans.
low latencystructured outputmemory-aware
PACK 04
Trading
Market-state interpretation, strategy rules, risk controls, execution preparation, and strict action constraints.
risk-gatedmarket toolsapproval-first
PACK 05
Developer
Repository navigation, debugging, patch generation, testing, structured tool use, and codebase context.
repo contexttool callspatch output
PACK 06
Custom / Enterprise
Private specialist packs tuned to organization-specific workflows, data, tools, policies, and deployment requirements.
private datadedicated evalsmanaged serving
MODEL PIPELINE
A repeatable path from open base model to specialist agent.
Most early PandOS packs should be produced through fine-tuning, distillation, and domain-specific evaluation rather than training full foundation models from scratch.
01
Open base
02
Domain dataset
03
SFT / distill
04
Task evals
05
Quantize
06
PandOS pack
DISTRIBUTION TIERS
One specialist family.
Multiple compute targets.
The same purpose-built intelligence can ship as a free compact local download, a higher-quality workstation build, or managed cloud inference for enterprise and external agent marketplaces.
Q4
Local
Mainstream consumer hardware
Q6 / Q8
Local+
Stronger local machines
FP16 / BF16
Server
Workstation or private deployment
Managed
Cloud
Enterprise and marketplaces
PACK CONTRACT
More than weights.
A Model Pack should ship the intelligence and the operating contract that makes it useful as an agent.
Model weights + tokenizer
System policy
Tool definitions
Context templates
Task-specific evals
Runtime configuration
FLAGSHIP PACK