Private AI • verified intelligence • measurable compute
The model thinks.
KESPA knows.
A private second brain that learns what it can prove. KESPA combines local reasoning, trusted knowledge, evidence-backed research, memory, provenance, and adaptive tools while keeping private context under customer control by default.
public telemetry
Phase: Controlled validation
RTX 3070 8GB
Qwen2.5-Coder 7B
0
Live retrieval cards
75
Active public trusted knowledge
0
Manual training events
3
Research releases received
76
Canonical NexLedger attestations
21
Tracked assistant responses
Live KESPA pulse
A platform that updates as it learns.
These counters are not hand-entered marketing numbers. The page refreshes from KESPA's tracked application state, knowledge lifecycle, research releases, usage telemetry, and canonical ledger proofs.
Latest activity
auto refresh
Intelligence lifecycle
Latest trusted knowledge
Waiting for live state…
Knowledge promotion activity will appear here.
Latest manual training
Waiting for live state…
Training lifecycle activity will appear here.
Latest NexLedger proof
Waiting for live state…
Canonical proof activity will appear here.
Research pipeline
From gap to reusable intelligence.
Pending knowledge candidates0
System-policy approvals0
Natural research requests pending0
Delivered research requests0
Public research entries0
Public research artifacts0
Trust boundary:
background model output does not become trusted knowledge automatically.
Research must still pass the evidence, verification, policy, promotion, and indexing lifecycle.
KESPA Community
The people actually using it.
Live community rankings are calculated from tracked product usage with benchmark traffic excluded. The logged-out homepage only identifies users who explicitly enabled public profile sharing.
Loading community…
Live KESPA Score
Research thesis
Can continuously verified external intelligence allow smaller local models to achieve materially greater useful capability per unit of recurring compute — while preserving privacy, provenance, and model interchangeability?
Current research checkpoint
Measured work, not a static demo.
KESPA publishes positive results, negative results, methodology corrections, lifecycle work, and provenance as the system evolves.
Latest verified research release
Key Facts About Life Purpose
What is your purpose?
1 verified claims
2 evidence records
2 source organizations
forge-release-bf03200aaaba424029d46688
Canonical benchmark / V1 finding
Verified intelligence beat both raw inference and legacy orchestration.
On the locked 54-case A/B/C benchmark, the verified-retrieval arm reached 54.22 quality at 1.675 s mean latency and 0.1085 Wh mean GPU energy.
B quality 54.22
Hit@1 54/54
Gold recall 79.17%
Adaptive compute / D v001
Negative results stay visible.
D v001 preserved strong quality and acceptable compute, but failed its preregistered Hit@1, recall, and escalation targets. The primary success criterion remains recorded as a failure.
Quality 55.2586
Latency 2.443 s
Primary success: FAIL
Public research archive
Research archive loading…
Public experiments, benchmarks, technical notes, and provenance are synchronized from an immutable repository revision.
0 entries
0 artifacts
SUPPORT KESPA
Help us keep building the system.
KESPA is independently developed by NexLabs Studios. Direct support helps fund compute, research, infrastructure, and the engineering required to keep pushing the project forward.
Support development directly.
Make a one-time contribution through Stripe-hosted Checkout and choose exactly how much you want to put behind KESPA.
Support KESPAPut infrastructure or resources behind it.
We welcome hardware, compute credits, research collaboration, infrastructure sponsorship, and strategic partnerships.
PartnershipsInterested in NexLabs Studios or KESPA?
Request project information, research material, technical documentation, and a direct conversation with NexLabs Studios.
Investor informationSystem architecture
Intelligence lives outside the model weights.
Models are replaceable compute. KESPA owns the durable intelligence layer: evidence, trusted knowledge, memory, provenance, policy, permissions, retrieval, and compute allocation.
01 / Knowledge lifecycle
Evidence before trust.
Raw information becomes candidate claims, then passes evidence binding, verification, policy, conflict controls, promotion, indexing, and provenance before becoming durable trusted intelligence.
Evidence→Claims→Verify→Policy→Trusted→Retrieve
02 / Model
The model reasons.
KESPA supplies relevant trusted context. The local model spends compute on reasoning instead of trying to memorize the whole world.
03 / Privacy
Private by default.
Private context does not become public research or public Chroma content merely because a model touched it.
04 / Research plane
Repair genuine knowledge gaps.
Natural research can discover public sources, fetch them directly, qualify evidence, verify claims, and return a frozen release.
05 / Provenance
MySQL → Chroma → NexLedger.
MySQL remains authoritative, Chroma remains rebuildable retrieval infrastructure, and NexLedger provides canonical tamper-evident attestations.
06 / Current roadmap
Make KESPA useful, then make it broader.
NOW
Closed-alpha readiness
Frontend • reliability • telemetry • privacy
→
NEXT
Trusted-friend alpha
3–10 real users / real abuse data
→
BUILD
Evidence Store MVP
TXT • PDF • image • spreadsheet
→
GROW
1K trusted claims
Then freeze a longitudinal benchmark
Current configured phase:
Controlled validation