BABELRESEARCHZeakryy suite ↗

A SPACE FOR INQUIRY v0.1.1

A question is
only the beginning.

Follow the evidence. Keep the uncertainty.
A research workspace built to show its work.

Explore a verified run ↗See what works ↓

Development report · Checked September 26, 2026
A dated local audit, not a live service monitor.

SYSTEM ASSESSMENTPartially operational

The foundation works.
The AI layer is offline.

Local evidence runs and library retrieval passed fresh checks. AI synthesis needs its local services started and retested.

Evidence pipelineVerified
Knowledge retrievalVerified
AI synthesis & web researchUnavailable now
Execution locationLocal PC
No autonomous research service is currently hosted here.
26/26

Researcher tests passed

3 → 9

Sources → exact excerpts

108

Topics in the library

22/22

Role contexts retrieved

01 / CAPABILITIES

What it can do today.

Four research modes. Different levels of readiness.
Tests and live runs are identified separately.

01Verified locally

Local evidence research

Search local documents, preserve exact excerpts, connect claims to evidence IDs, and generate a reviewable packet.

  • Markdown, text, JSON and CSV inputs
  • Review gates, event log and file hashes
  • No model or API key required
Limit Deterministic retrieval. It identifies relevant material; it does not perform model reasoning.
02Verified locally

Role-guided library retrieval

Retrieve relevant training and rehabilitation topics with the claims, source records and prerequisites a role needs.

  • 18 domains · 147 sources · 160 claims
  • Read-only SQLite full-text search
  • All 22 role contexts checked
Limit Curated paraphrases, not exact paper quotations. Search rank is not evidence strength.
03Services offline

Local AI synthesis

The AI.OS adapter can ask a local model to organize supplied evidence and propose claims over known evidence IDs.

  • Evidence stays inside the local stack
  • Unknown evidence IDs are rejected
  • Adapter tests pass; fresh live run pending
Blocker AI.OS and Ollama did not respond at audit time. Current model output quality is unverified.
04Not verified live

Open-web research

An installed GPT Researcher adapter has a profile for DuckDuckGo discovery with local Ollama synthesis and embeddings.

  • Keyless profile exists in the package
  • Questions and search terms leave the PC
  • External output remains candidate synthesis
Gap No fresh web run completed. The adapter does not independently verify pages or extract quote-level evidence.

02 / INSIDE A REAL RUN

From a question to a paper trail.

Download audited snapshot ↓
QUESTION

What does the local evidence say about identity persistence?

Fresh offline run · Bundled example documents · 152 ms pipeline time · No language model

Question→Sources→Excerpts→Claims→Review→Packet

PACKET EXPLORER

Sanitized example packet.
Private paths and raw workspaces are excluded.

Loading the audited packet…

Exact excerpts prove what these example files contain. Passing review gates and matching hashes do not establish scientific truth or independent corroboration.

03 / KNOWLEDGE LIBRARY

A map you can actually navigate.

Browse the published topic index.
Retrieval examples below come from the local engine.

Loading topics…Browser title filter · No AI request

Library integrity needs attention. Four notes/Canvas files differ from the saved manifest. SQLite retrieval passed, but those edits have not been reconciled; they were preserved.

RETRIEVAL EXAMPLE

Endurance coach

These saved results include prerequisites and connected claims. They are not a personalized training plan.

04 / WHERE IT RUNS

Research on the PC.
Reports on the web.

Local execution · Cloudflare publishing

One place to do the work

The researcher, model, library and full evidence packets stay on the local PC. AI research requires that PC to be awake and its model service running.

Cloudflare serves this website and explicitly reviewed snapshots. The page stays accessible when the PC is off; it never calls into the local researcher or model.

CURRENT ARCHITECTURE

Run locally → review the public snapshot → publish a frozen release → browse from anywhere.

CLEAR RESPONSIBILITIES

Local PCSource acquisition, model inference, library retrieval, private packets and review.
CloudflareStatic webpage, approved sample evidence and dated status snapshots. No model inference or research-job API.
OracleNot used in this setup. The inspected server remains available for a future workload that actually needs it.

Static asset requests and storage have no additional charge under Cloudflare’s published Workers Static Assets pricing. No paid model API or new Oracle allocation is required.

Static asset pricing ↗
ActivityWhereWhen the PC is off
Browse this report and published examplesCloudflareAvailable; shows the last reviewed snapshot
Run research and use the local modelLocal PCUnavailable until the PC and model are running
Submit research from a phoneFuture optionNo job queue is implemented in this release

05 / WHAT COMES NEXT

Progress, with receipts.

The pipeline is a working foundation.
The researcher and webpage overhaul comes next.

Read the full status report ↓
  1. ✓

    Evidence foundation

    Offline packets, citations, review gates, hashes and library retrieval verified.

  2. 2

    Restore and evaluate the AI layer

    Start local services; run fresh synthesis and keyless web acceptance with recorded quality and timing.

  3. 3

    Close the web evidence gap

    Retrieve source text independently and attach verified excerpts before promoting external synthesis.

  4. 4

    Overhaul the researcher and webpage

    Build on this baseline. Improve research quality and the interface while keeping execution local and publication deliberate.