AI systems / operational reliability
AI Agent Systems
Browser-based agent operations and continuity
Local orchestration experiments around browser AI sessions, context continuity, deduplicated actions, and project-specific agent workflows.
The problem
Browser-based AI agents have session boundaries, mutable external state, and action retries. Without continuity and evidence of execution, long-running work becomes fragile.
What I built
An evolving set of local orchestration tools and experiments for keeping multiple project agents separated while tracking progress and recovering context between browser sessions.
Engineering decisions
- Separate account identity, project scope, and execution authority.
- Distinguish requested actions from externally confirmed results.
- Preserve deduplication and opt-out rules while avoiding unwarranted global locks.
- Record observable test results instead of treating an architecture proposal as implemented behaviour.
Further reading
Read the browser-harness engineering write-up. This is a changing development record, not a guarantee that every discussed feature is deployed.
Evidence and external reach
Independent references
These records describe a publication, directory inclusion or data-use entry. None implies institutional endorsement.
No third-party source recorded for this project yet. Its absence is not a judgement about the project.