Roadmap
Product roadmap
What Tether is building next across capture, recovery, desktop clients, integrations, and release quality.
Roadmap principle
Tether is not trying to become a hosted agent workspace. The product direction is a local execution debugger for AI coding agents: capture the real run, preserve exact evidence, show the failure path, and make recovery faster.
- Prefer local execution evidence over generic analytics.
- Keep prompt, action, file diff, command, test, and recovery data connected.
- Ship debugger controls only when the source can support them honestly.
- Avoid fake replay promises for events that only have log evidence.
Now
| Area | Focus | Outcome |
|---|---|---|
| Execution capture | Local source logs, proxy-captured requests, local trace storage, and normalized graph nodes. | A developer can see the prompt, model request, response, metadata, and source context for a local agent run. |
| Desktop debuggers | macOS and Linux clients with graph selection, inspector panes, settings, local proxy control, and source-log ingestion. | The desktop apps are the primary local debugging surfaces. |
| Recovery evidence | Replay for supported proxy-captured requests and rollback context for file-changing nodes. | A failed branch has enough evidence for targeted recovery instead of blind reruns. |
| Public docs | Architecture, supported agents, commands, privacy, release workflow, and API reference. | The site explains the real product shape instead of presenting Tether as a generic observability tool. |
Next
| Area | Planned work | Why it matters |
|---|---|---|
| Capture wrapper | Harden tether capture -- <agent command> across local coding-agent workflows. | The wrapper is the cleanest path to full prompt, process, shell, file, and test attribution. |
| Failure-first graph | Make failed commands, failed tests, changed files, and stale downstream branches easier to scan. | The graph should answer where the run broke before it asks users to inspect every node. |
| Replay boundaries | Make replay availability explicit per source, request type, and provider route. | A debugger loses trust if it shows controls that cannot actually recover the selected node. |
| Adapter coverage | Deepen supported CLI, framework, OpenAI-compatible, and custom source ingestion. | Tether should follow the agent stack developers already use instead of forcing a new runtime. |
Later
| Area | Direction | Constraint |
|---|---|---|
| Windows client | Build a desktop client around the same local proxy and execution graph contract. | Do not add placeholder runtime code until it can build and run. |
| Team evidence export | Export traces, failure reports, and recovery context for issue trackers, pull requests, or incident notes. | Exports must avoid leaking secrets and must preserve exact file and command attribution. |
| Release quality | Tighten signing, notarization, update checks, smoke tests, and packaged helper verification. | Distribution quality has to catch up before broad public release. |
Not on the roadmap
- A hosted web IDE as the main product surface.
- Generic LLM observability dashboards detached from local file, command, and test evidence.
- Provider-specific lock-in that makes Tether useful only for one model vendor.
- Replay buttons for sources that cannot safely replay the selected execution boundary.