The control plane for your coding agents.
Relay isn't another dashboard — it's where you run the work. Plan tasks on a Kanban, launch real terminal sessions straight from a card, and drive every agent — .claude, .codex, .cursor, .gemini — from one board. Cost and session tracking come standard. All on your machine, zero setup.
- // Local-first
- // Every agent
- // Zero telemetry
// features
A control room for your /agents.
Six primitives, one binary. Everything you need to plan, launch, and drive the agents you already run — observability is just the part that comes for free.
Kanban Deploy
Plan the work, then run it. Drag a card → Relay opens a new terminal and launches your CLI agent with the task prompt already seeded. Backlog, in-flight, and done, all on your machine.
Native Terminals
No custom runtime, no sandboxed shell. Relay spawns a real session in your own terminal, running the same `claude` / `codex` / `cursor-agent` command you'd type by hand.
Every Agent, One Board
Point Relay at ~/ and it discovers every .claude, .codex, .cursor, .gemini session on disk. No SDK, no instrumentation, no manual wiring — even for agents you never set up to be watched.
Session Replay
Rewind any agent run. Diff prompts, tool calls, and file writes. Fork from any step to try a different path.
Spend Tracking
Every token metered locally. Per-agent, per-repo, per-day. Set caps and get shell alerts before the bill spikes.
Live Observability
Watch prompts, tool calls, and file diffs stream in real time. Filter by agent, repo, or exit code.
// local-first manifesto
Data stays on your machine.
Cloud dashboards ship your prompts, code, and API keys to a stranger's S3 bucket. Relay refuses. Every byte lives beside your source, encrypted at rest, invisible to us.
On-disk only
Sessions, spend, config — persisted to ~/.relay. Never leaves your box.
Zero telemetry
No pings, no analytics, no phone-home. Sniff the packets, we dare you.
Your keys, your rules
Relay reads existing CLI auth. We never see, store, or proxy your API keys.
// vs the rest
The receipts.
Everyone tracks tokens now. The last two rows are table stakes — the rest is the part only Relay does. Feature-for-feature against cloud agent platforms and LLM observability tools.
| Capability | RELAYnative · local | Cloud Agentsdevin · replit | LLM Observabilitylangfuse · helicone |
|---|---|---|---|
| Kanban → real terminal spawn | ✓ | — | — |
| Multi-CLI support (claude/codex/cursor/gemini) | ✓ | — | — |
| Runs entirely on your machine | ✓ | — | — |
| Uses your existing CLI auth | ✓ | — | — |
| Zero telemetry, zero cloud | ✓ | — | — |
| Per-agent spend & session tracking | ✓ | ✓ | ✓ |
| Live prompt & tool-call observability | ✓ | ✓ | ✓ |