PROTEAN INTELLIGENCE SYSTEM · OPEN-SOURCE PORTFOLIO
An AI team with defined roles. A system you can inspect.
Protean Intelligence System gives AI assistants defined roles, shared project notes, handoff rules, and review steps. ASKA uses it to organize its own agent work and offers setup, training, and support for your deployment. The public kit is an ASKA open-source project—not a client case study.
v1.6.0 · current releaseApache-2.0 · kits layer7 · AI roles30 project workspaces
THE SCALE PEOPLE DO NOT SEE · ONE MACHINE, FIVE LAYERS
This is not one chatbot with a few prompts around it.
The interface sits above several operating layers: agent roles, saved project state, model selection, optional local preprocessing, and review. The inventory below describes ASKA’s documented setup, not a performance benchmark or a live counter.
7AI roles
COORDINATED PROTEAN SYSTEM PROFILES
Fixed role lanes with defined mandates. Work moves through choreographed handoffs, not open chat.
↓every profile is bound to project state, not to a session
30local project workspaces
LOCAL PROJECT WORKSPACES
Durable state, routed into bounded context. Every source file carries a manifest verdict; an unclassified file fails the build.
FIXED ROLE LANESMANIFESTSLICENSE CHECKSSTAGING ISOLATION
↓knowledge routes work to the smallest capable model
22cloud models catalogued for routing
GOVERNED MODEL STACK
Four tiers, one routing policy: frontier cloud, local quantized midrange, local specialists, and experimental local variants — including abliterated builds where appropriate, behind evaluation, provenance, licensing, and safety gates.
↓local gates screen work before it reaches a frontier call
3automatic local functions
LOCAL SPECIALIST GATES
Redact screens approved outbound text and holds address, numeric, or uncertain findings for review. Gist supplies topic hints only — never a decision. Title drafts artifact metadata.
REDACTGISTTITLE
LOCAL REDACTIONDEPLOYMENT LOCKSFIVE-MINUTE SUPERVISION
↓nothing is done until the served artifact proves it
HUMAN REVIEW BOUNDARY
Producer / verifier separation — the agent that produces an artifact never passes it.
Served-byte verification — checks run against the bytes served, not the agent’s report.
Fresh-clone tests, phase gates, and recovery receipts — a stranger can reproduce it, nothing ships past an unsigned review, every restart leaves a paper trail.
Remove any layer and the others stop being an operating system.
01 / WHAT PROTEAN SYSTEM IS
A governed team, instantiated from one parameter file.
Protean Intelligence System defines role lanes with identity archetypes so each agent has a fixed mandate, choreography for how orchestrators and workers hand work to each other, governance rules and phase gates so nothing ships past an unsigned review, knowledge routing that keeps persistent memory small enough to fit a bounded context, and provenance enforcement where every source file carries a manifest verdict and an unclassified file fails the build.
ROLE LANES
Identity archetypes, fixed mandates
Each agent has one lane with a defined role, so a team is a set of distinct mandates rather than a crowd of interchangeable prompts.
ORCHESTRATION
Choreography with defined hand-offs
Orchestrators and workers pass work through explicit choreography, so supervision is a designed stage, not an afterthought.
GOVERNANCE
Rules and phase gates
Phase-gated orchestration means nothing ships past an unsigned review. Disciplined by structure, not by vigilance.
STATE & MEMORY
Durable state and bounded context
Knowledge routing keeps persistent memory small enough to fit a bounded context, so a long-running team does not drift into a bloated, contradictory past.
PROVENANCE
Served-truth verification
Every source file carries a manifest verdict and an unclassified file fails the build. What is claimed is what is served, and what is served is what is checked.
GENERATOR
One file assembles a team
A generator is the only assembly path: it instantiates a full configured team from a single parameter file, and the repository ships a frozen-clone proof point so a stranger can reproduce the instantiation from a clean clone.
Built on the Hermes engine. The kits layer is Apache-2.0. Current release: v1.6.0.
02 / LOCAL PREPROCESSING
Small specialist models around the team, not instead of it.
Protean Intelligence System can use a guarded local preprocessing layer when a narrow task does not need a large reasoning model. Redact can prepare approved outbound text and hold address or uncertain findings for review. Gist can provide cheap topic hints for bulk intake. Title can draft metadata for artifacts and transcripts.
PRIVACY
Local first-pass redaction
Sensitive text can be screened on-device before an approved outbound step. The filter is not legal anonymization; uncertain and address-related findings stop for review.
EFFICIENCY
Cheap hints and metadata
Topic hints and draft titles reduce repetitive work before Protean Intelligence System performs interpretation, judgment, and final synthesis.
The layer is automatic when the workflow benefits from it, but bounded: it does not replace Protean Intelligence System reasoning, make high-stakes decisions, route solely on a small model, or download media models during active work. Model licensing remains separate from the Apache-2.0 kits layer.
03 / THE MODEL STACK
Different models. Different jobs. One governed system.
Protean Intelligence System is designed to route across a spectrum rather than worship a single model. Frontier cloud models handle difficult synthesis and broad reasoning when the approved environment allows it. Local quantized midrange models handle repeatable work near the data. Specialized local models handle narrow transformations. Experimental local variants, including abliterated builds where appropriate, are treated as bounded technical options — not as magic — and remain subject to evaluation, provenance, licensing, and safety gates.
FRONTIER CLOUD
deep synthesis · architecture · difficult judgment
↓ routed by task, privacy, cost, and capability
LOCAL QUANTIZED MIDRANGE
private work · repeatable analysis · lower-latency execution
The point is not maximum model count. The point is matching capability to the job while preserving a reviewable chain of custody. Models are replaceable. Routing policy, role separation, provenance, durable state, and verification are the durable asset.
04 / HOW ASKA USES IT
Governed operating practice, not ad-hoc chat.
For ASKA, Protean Intelligence System is the shared layer underneath the firm’s own agent work. Agent teams run on it as governed operations — with role lanes, phase gates, durable state, and provenance — rather than as a scatter of one-off conversations. It is open source on purpose: the engine stays upstream and unmodified, the kits layer is Apache-2.0, and vertical packs remain proprietary by contract. What is public is the general method, not any client deployment.
05 / THE BOUNDARY
What this kit is, and is not.
ASKA keeps the honest read on the page with the project. Three lines hold it.
Portfolio, not a case study. Protean Intelligence System is an open-source ASKA portfolio project. It is not a client case study, and no client name, engagement, or outcome is attached to it.
Breadth, not vertical proof. The public kit is the operating layer ASKA uses for multi-agent work. It does not by itself prove performance in any specific vertical; vertical deployments run through proprietary pack parameters and services that are not part of the public repository.
“Largest” defined honestly. ASKA calls it the largest open-source portfolio project by the breadth of the system’s surface — identity, orchestration, governance, memory routing, build gates, and the generator — and by its role as the shared layer under ASKA’s agent work. Not by star counts, file counts, or lines of code.
06 / PROTEAN SYSTEM AS A SERVICE
The kit is open source. The service is what ASKA does around your deployment.
Protean Intelligence System is the open-core operating layer, released as an open-source ASKA portfolio project. Running it for real — in your environment, with your data and your models — is professional work. That work is the ASKA Protean Intelligence System service: setup, deployment, training, and the ongoing operation of a governed agent team. Two ways to engage, scoped to your environment during consultation.
ONE-TIME
Setup, Deployment & Training
$9,300 USD
One-time setup, local deployment & training, when the client provides the local hardware. Training is included.
Environment design and standing up your Protean Intelligence System deployment, agreed to your scope.
Onsite setup is available — ASKA installs in your environment.
Deploys on client hardware, ASKA-supplied hardware, or hosted/remote where appropriate.
Local operation can be configured with suitable installed models and hardware. Cloud services and external tools remain optional deployment choices.
Onboarding and training so your team runs it as a governed operation, not ad-hoc chat.
Ongoing operation after deployment, scoped to your environment.
Monthly Protean Intelligence System upgrades, reviewed for upgrade impact and compatibility before being applied.
AI/agentic-technology newsletter covering relevant developments and practical implications.
Model retuning and recommendations toward the best, most efficient models for your needs — subject to your approved environment, provider, budget, and security constraints.
Quarterly recalibration of the Protean Intelligence System structure, roles, workflows, and governance to your changing needs.
Bug fixing and troubleshooting.
Maintenance and operational support across your deployment.
Configuration and prompt/skill tuning within the agreed deployment scope.
Backup/recovery and health-check guidance for local deployments.
Quarterly service review with a prioritized improvement backlog.
Security and provenance hygiene checks on Protean Intelligence System configuration and adopted external components, within the agreed scope.
Light usage/performance review and recommendations.
The kit can be deployed locally on suitable client hardware. If ASKA supplies the hardware instead, that is separate from the setup fee and is quoted according to the performance you need — hardware scope is set by required performance, not a single published figure. Service scope is bounded by your deployment, hardware, provider access, security policy, and the agreed support terms. ASKA does not guarantee uptime, response times, savings, or model performance.
Start with a free consultation call and Protean Intelligence System product demo.
07 / THE PROJECT
Open the kit, or read the code.
v1.6.0 · Apache-2.0 (kits layer). The full public site walks the operating layer end to end; the repository is the same work as source.
ASKA DIGITAL