| Filename | Latest commit message | Latest commit date |
|---|---|---|
| README.md | ||
web-chat
A planned cross-platform successor to ego-chat that preserves its functionality and supports durable discussions and automation until a question is settled.
Status
Research and planning only. This repository currently contains this README; no runtime, feature parity, migration, or browser integration is implemented or verified.
The reference baseline is x/ego-chat at commit a8beaa58f4806c571639e2008f4b1efd0073531b (v0.2.47). The original repository remains unchanged.
Feature-parity contract
Before implementation, inventory every baseline capability and map it to an acceptance test. Include user-facing workflows, commands and APIs, configuration, conversation state, automation, integrations, recovery behavior, and platform-specific behavior. Record each item as unverified, implemented, tested, or explicitly deferred with an agreed reason. Feature parity is a release gate, not a claim made by creating this repository.
Preserve existing behavior through compatibility boundaries and migration tests. Introduce the new discussion engine alongside the compatibility layer rather than silently replacing supported workflows.
Proposed architecture to evaluate
- Portable core: Keep conversation, task, policy, and persistence logic independent of operating-system and browser details.
- Browser adapters: Define a capability-based interface for browser discovery, session attachment, navigation, observation, interaction, downloads, and recovery. Evaluate adapters for supported Chromium/CDP and Playwright environments, with an explicit operating-system support matrix. Keep browser profiles and credentials isolated; do not copy them between environments.
- Durable execution: Persist discussion events, checkpoints, pending actions, and resumable execution state. Use stable action identifiers and idempotency checks so retries do not duplicate external effects.
- Human control: Support interruption, pause/resume, amended goals, approval gates, and explicit cancellation. Recovery must preserve pending approvals and user intent.
- Discussion convergence: Track the goal, constraints, evidence, unresolved questions, candidate conclusions, and the reason for stopping. Continue useful investigation while issues remain; distinguish settled, blocked, awaiting input, paused, and cancelled states. Detect repetitive or unproductive loops and ask for direction instead of running indefinitely. Resource limits must produce a visible pause or escalation, not a false claim of resolution.
These are design candidates, not implemented components or finalized technology choices.
Research work
- Audit ego-chat at the pinned baseline and build the feature/acceptance matrix.
- Compare open-source patterns for persistent workflows, checkpointing, human-in-the-loop control, browser adapters, and inspectable agent execution. Assess fit, maintenance, operational complexity, testability, and licensing before adopting dependencies.
- Locate and review the self-hosted skills repository. Assess reusable instructions and packaging separately from executable installers; review provenance, permissions, and compatibility before installation.
- Choose the smallest architecture that meets the parity contract, then validate it with a thin end-to-end slice and restart/retry tests.
Repository boundaries
This repository is private. No source code, workflows, credentials, or external integrations have been imported. CI, runners, deployment, and skill installation require deliberate setup after review. No project license has been selected; audit the source and dependency licenses before code reuse or redistribution.