We think with AI and create new experiences through creativity. From personal reflection with OpenAlarm to portfolio-based research with OpenStockAI, explore how our ideas become products.
01 / Available now
OpenAlarm
Your patterns. Your feelings. A reply that connects.
OpenAlarm considers recurring patterns and emotional context in your diary to shape a personal AI reply. Explore suggestions based on alarm use, revisit past entries and trace place-linked plans on a footprint map.
Records are handled within the user's account scope. Diary input is separated from instructions, and replies containing secrets or internal information are rejected before storage with a security alert.
OpenStockAI puts a question in portfolio context, routes it through the appropriate research path, and organizes the response around evidence, dates and risk. It is research support, not a return promise.
Research agents work alongside a security team. Checks identify exfiltration and privilege-bypass signals, connect request blocking to redacted audit records, and support review.
AI that divides the work.
Boundaries that stay in place.
Orchestrators connect tools and agents, while product-specific safeguards check access, inputs and outputs. Explore where detection, blocking and limits on sensitive-data transmission take place.
Research agents work alongside a security team. Checks identify exfiltration and privilege-bypass signals, connect request blocking to redacted audit records, and support review.
Architecture walkthrough · not live security telemetry
OpenStockAI / 01
Input
Question · portfolio context
Selected work
01Research tools
02Specialist agents
03Synthesis & review
Security checks at request time
AI input protection and training policy
Google API user data is blocked from designated external-model inputs. Supported paths disable response storage, while security analysis and records redact sensitive information.
Read the protection scope
The security team combines server rules and optional AI-assisted review. Detected signals and operating settings determine logging, request blocking and temporary blocks.
A harness applies safeguards around AI execution. Transmission controls, redaction and output checks reduce exposure of sensitive information; provider retention and training policies are managed separately.
Inside the work / 03
Behind every result, a process.
Choose a product and follow its process. See what each stage receives, what it does and what it produces.
OpenDataLabs / Documents & schedules
From reading documents to managing what comes next.
Connect outbound checks, OCR re-reading, Jev verification, reference comparison and human review. Bring shipping, delivery and payment deadlines from documents and notes into a schedule with source evidence and completion tracking.
Workflow walkthrough · On narrow screens, scroll horizontally to explore.
Workflow walkthrough / 01 — 13
OpenDataLabs
Document input
Input
PDF · images · spreadsheets
Work
Read content and source positions using the appropriate parser.
Output
Parsed content
OpenDataLabs
OCR & re-reading
Input
Images · scans · handwriting
Work
Retry low-confidence reads with preprocessing profiles and compare measured results.
Output
Text, coordinates and OCR confidence
OpenDataLabs
Extract & organize
Input
Parsed content
Work
Extract and normalize fields, marking uncertain values.
Output
Fields linked to their source
OpenDataLabs
Privacy Harness
Input
Task-specific payload
Work
Validate allowlisted fields, lengths and forbidden keys/values. Block violating requests; exclude digit-containing fragments from text-repair tasks.
Output
Validated minimal input
OpenDataLabs
Jev verification
Input
Allowed fragments · alternatives
Work
When external AI is allowed and connected, use providers such as Jev to verify word and spacing candidates.
Output
Candidate selections and verification
OpenDataLabs
Correction checks
Input
Candidates · verification · source
Work
Apply internal checks to correction proposals and preserve transformation history.
Output
Corrected fields and history
OpenDataLabs
Compare records
Input
Fields · suppliers · POs · receipts
Work
Resolve document fields against reference records.
Output
Comparison results
OpenDataLabs
Check rules
Input
Fields and references
Work
Check quantities, amounts and required fields.
Output
Individual check outcomes
OpenDataLabs
Result & evidence
Input
Checks · source references
Work
Keep the machine decision, evidence and review items.
Output
Machine result
OpenDataLabs
Human review
Input
Review items and source
Work
A person corrects or confirms the fields with a recorded history.
Output
Human confirmation
OpenDataLabs
Record confirmation
Input
Corrections · confirmation
Work
Preserve the original machine decision separately from human confirmation.
Output
A result with clear provenance
OpenDataLabs
Connect schedules
Input
Document dates · item deadlines · notes
Work
Collect shipping, delivery and payment deadlines into source-linked events. Leave ambiguous dates unresolved for review.
Output
Source-linked schedule
OpenDataLabs
Track & follow up
Input
Events · actions · receipt evidence
Work
Review approaching and overdue events, mark them handled or reopen them. Delivery and shipping events also use linked receipt evidence.
Output
Follow-up work and completion status
Connected stages
In development. External AI requires an allowed connected mode. Optional image assistance has a separate permitted path; this is not a claim that source content never leaves. Detected resident-ID patterns block that image request. Accuracy improvements require separate evaluation.
In the works
OpenDataLabs
Connect verified documents and handwriting to shipping, delivery and payment schedules. Track deadlines and completion with source evidence and human review. In development.
A Privacy Harness checks task-specific allowlists before information reaches an external model. It limits unnecessary source and identifying data and blocks policy-violating requests.
