OBSERVABILITY
Measure the real state
Read-only health snapshots, resource counters, bounded logs and service checks. Status labels distinguish online, offline, degraded and unknown.
OS29 · LOCAL-FIRST OPERATIONAL INTELLIGENCE
Build local tools.
Verify what works.
Keep evidence.
OS29 brings local observations, trace lineage, redaction, capability state, mission capsules and bounded MCP tools into one evidence-first workflow. Declared, configured, running and tested behavior stay distinct, and unfinished capabilities remain clearly marked.
ANA MAX / OS29LOCAL TOOLING + EVIDENCE
29TRACE
TEST EVIDENCERECORDED
GUARDRAILSMCP / LOCAL
ENGINEERING PRINCIPLEEvidence first.
Engineering focus areas
06 PRACTICAL DISCIPLINESAUTOMATION
Connect tools to workflows
Python and PowerShell entry points, MCP/JSON-RPC routing and local utilities with explicit execution boundaries.
MEMORY & RETRIEVAL
Persist useful evidence
SQLite persistence, linked trace events, mission capsules and local FTS5/BM25 search. Candidate notes remain separate from validated lessons.
QA & TESTING
Prove a change
Pytest suites, syntax and contract checks, bounded smoke tests, and reports that keep skips and missing prerequisites visible.
GUARDRAILS
Scope every operation
Default-deny target and operation policies, bounded inputs, loopback-only probes and explicit approval boundaries.
ENGINEERING DISCIPLINE
Make changes maintainable
Backups, incremental changes, compatibility checks and evidence-based status labels reduce regressions across versions.
LIVE-RUN READINESS
Missing prerequisites stay visible.
This is a static portfolio, not a remote controller. It cannot inspect a visitor's device, start OS29, or call a local MCP endpoint. The readiness items below are not live health checks.
LOCAL RUNTIME
Runtime and endpoint
Start the intended checkout and Python environment, confirm required packages/configuration, then verify the loopback MCP endpoint from the local operator session.
NOT CHECKED BY THIS PAGEACCESS CONTROL
Scope and approval
Select the intended tool profile; check default-deny target and operation rules, loopback binding, and explicit approval boundaries before any sensitive operation.
NOT CHECKED BY THIS PAGECONTROLLED DIAGNOSTICS
Frida / MITM readiness
The project includes Frida/MITM-oriented diagnostic paths for controlled failure visibility. They require an authorized app/device/process, compatible tooling, a defined scope and, for proxy testing, the correct proxy/certificate setup. The public page launches no hook or interception.
TARGET SETUP NOT VERIFIEDEVIDENCE & RECOVERY
Reproduce, fix, verify
For a controlled failure: capture a redacted error trace, run only the permitted retry/recovery path, rerun the relevant test and report the outcome against evidence. The replay's recovery line is simulated.
NO LIVE TEST RUNSESSION REPLAY · NOT LIVE
One intent,
shown as a replay.
A scripted walkthrough of the intended flow. The events below are illustrative; this public page does not execute commands, contact MCP or verify a recovery.
ARCHITECTPokee-Isaac
ORCHESTRATORAntigravity
OPTIONAL AI ASSISTANTSChatGPT · OpenAI Codex · Devin
These are example roles, not proof of authenticated or currently running connections. Availability depends on the actual local configuration.
Ready. Select Play to step through this scripted example.
A traceable execution path.
THE EXECUTION PATH01 / REQUEST
Model or operator
A request is planned and scoped.
Secrets and approvals stay outside the page.
02 / ROUTE
OS29 / LOCAL RUNTIME
Local tool routing and an evidence layer.
Scoped calls, redaction and retrieval by build.
03 / VERIFY
Observable result
Bounded tool response.
Health, tests and evidence—not guessed success.
Next acceptance gates.
OS29 ACCEPTANCE GATESTrusted approval context
Complete the per-tool, per-operation authorization matrix and make approval context verifiable before enabling broader write operations.
Retrieval quality
Improve the current SQLite FTS5/BM25 baseline only with measured retrieval evaluations, redaction tests and source-linked memory evidence.
Complete a real evidence loop
Use a permitted local test target, reproduce a controlled failure, apply a small fix, retest and validate only the result linked to that evidence.
SELECTED ENGINEERING WORK
OS29 — TRACEABLE OPERATIONAL INTELLIGENCE
OS29 · TRACE & SCOPE
Recorded local baseline · 19 Oct 2026
Operational state you can inspect
Normalized trace events, parent lineage, redaction, explicit local target scope, incident records, mission capsules and a narrow MCP STDIO interface make the evidence path visible.
- 37 tests passed in the recorded OS29 local suite
- 14 bounded tools in the recorded MCP STDIO catalog
- Loopback-first network policy and empty Frida device/process allowlists by default
OS29 · MEMORY & TRUST
Lexical retrieval · evidence-linked
Memory without invented certainty
SQLite FTS5/BM25 retrieval spans events, incidents, mission capsules, capabilities and explicitly approved local text. Candidate experiences remain separate from validated lessons.
- 375 documents in the recorded local lexical index
- 0 validated experiences in that checkpoint; no real fix/retest lesson was promoted
- Semantic embeddings, reranking and a model-mediated reasoning loop remain unfinished
Counts are historical snapshots, not live counters. The 30-module roadmap is a plan, not a claim of 30 production-complete modules.
PUBLIC DEMO · PRIVACY BY DESIGN
Browser-only · no AI required
Live Profile
A visitor-facing demo personalizes the page with a name chosen by the visitor and shows browser-exposed hints only. It does not collect hidden OS data or upload the profile.
- Responsive layout and accessible live status
- Unsupported APIs show unavailable instead of invented values
- No telemetry API calls from the static site
37OS29 tests passed*
14Tools in the recorded STDIO catalog*
375Documents in the local lexical index*
0Validated experiences at checkpoint*
Evidence note. The figures marked with an asterisk are recorded in local test and completion reports dated 19 October 2026. They are not live counters. A passing test suite validates the tested surface—not every advertised tool or the entire roadmap. The public page is a portfolio, not a remote controller.
PROFESSIONAL PROFILE
Practical engineering. Honest evidence.
Self-directed developer focused on AI tooling, Windows automation and quality assurance. Builds Python and PowerShell utilities, connects local workflows through MCP/JSON-RPC, and tests behavior using bounded checks, repeatable suites and explicit status reporting.
The working principle is simple: a registered tool is not automatically a working tool; a successful smoke test is not a security certification; a planned module is not a finished feature.
PythonPowerShellHTML / CSS / JavaScriptMCP / JSON-RPCpytestSQLite / FTS5HTTP APIsGit / GitHubQA & diagnosticsLogging & redaction