WORK NOTES
Not just the output, but the decisions behind it.
A guide to demos and evaluation records from independent development.
Examples from independent development, not client deployment case studies.
Oathra
AI Meeting
AI Secure
These links describe independent work, not customer deployment case studies or third-party audits. See the source for the latest scope and conditions.
Source review date: 2026-09-15
Real app recordings and saved artifacts. Waiting is condensed.
The app needs local services and a configured model, on macOS 14 or later. The download alone is not a hosted service. A screenshot is optional and goes to the connected model only when you send a question. Previously Astra.
Prototype an internal AI workspace or an assistant using local and external models.
45-second recording with real Gemini responses, synthetic speech input, and some scripted confirmations.
Account-free tasks and the verified-email, time-limited voice trial are different experiences. External meeting connections require setup and evaluation.
Prototype voice-based training, task capture, or an AI character experience.
Transcript checks, simulation (recorded calls by the built-in agent, GPT-4o mini and Gemini Flash), and documented real-call development. The 100-real-call evaluation is not complete.
A spoken agreement does not prove a booking exists in a business system. Simulation and real-call evidence are separate. Real calls need configuration and incur costs.
Prototype a phone workflow, add completion checks to a voice agent, or design consent and human handoff.
An interactive synthetic case and reproducible single-rule versus correlation evaluation.
Snapshot analysis, not continuous monitoring or enforcement. It does not confirm exfiltration. Published evaluations use synthetic data, not production performance guarantees.
Prototype log ingestion and evidence-aware investigation. Not a replacement for production security controls.
Real-model work records and evaluation results including failures. The featured research replay ended partial.
A quality advantage over a single agent has not been established. Published comparisons include failures and provider constraints. Production acceptance is separate.
Prototype agent workflows with explicit roles, artifact checks, budgets, permissions, and human approval.
A film made by the tool itself, a real launch kit, and local verification records.
Single-operator alpha. Generating material is not publishing it. Real social-account publishing remains unverified; engagement and business outcomes are not guaranteed.
Design a recording-to-launch workflow, brand-specific templates, and approval steps.
A film made to explain Noa; not a recording of a live stream.
It is not a system that can never give an inappropriate reply. The conversation AI has usage fees, and running it uses electricity. A recording of a full live run has not been published.
Character-led streams, guidance in shops and venues, and prototypes for handling enquiries.
START A CONVERSATION
An unfinished idea is a good place to start. Tell us about the work, and what you want to try.