readme: lead with the competitive knockout — faster than Warp, local AI, governed, does their tricks too

Author: mdheller

README.md

@@ -2,7 +2,38 @@
 
 > They say the world was built on the back of a turtle. TurtleTerm carries the shell on its back.
 
-TurtleTerm is the SourceOS policy-aware, agent-addressable terminal workbench for trusted command execution, terminal receipts, agent delegation, and reproducible operator workflows.
+**The AI terminal that's faster than Warp, runs its AI fully local, and proves every agent action — then does all of Warp's tricks too.**
+
+TurtleTerm is a WezTerm-based terminal with a full AI agent fabric built in: inline command autosuggest, a self-hosted co-pilot, governed autonomous task execution, workflow libraries, and live session sharing — none of it cloud-locked, none of it metered.
+
+## Why TurtleTerm beats Warp
+
+| | TurtleTerm | Warp |
+|---|---|---|
+| **Core speed** | WezTerm / WebGPU (faster) | ~8ms input latency |
+| **AI backend** | Local (Ollama / Noetica) *or* Claude — your choice | Proprietary US cloud only |
+| **Cost to use AI** | $0, self-hosted, no meter | $20/mo, 1,500 credits, or BYOK |
+| **Telemetry** | None required | Must be ON to use AI on Free |
+| **Inline autosuggest** | ✅ local ghost text (~1ms history path) | ✅ cloud |
+| **Workflow library** | ✅ `turtle-drive`, versioned in *your* Gitea | ✅ Warp Drive (their cloud) |
+| **Live session sharing** | ✅ self-hosted over WezTerm mux | ✅ their cloud |
+| **Autonomous agent** | ✅ plan→execute→verify, **default-deny destructive** | ✅ agent mode |
+| **Replayable, attestable agent runs** | ✅ `ReasoningRun/Event/Receipt/ReplayPlan`, sealed tamper-evident | ❌ nothing in this class |
+| **Voice→shell · personas · AI coach · cost tracking** | ✅ | ❌ |
+| **Remote multiplexing** | ✅ native, no remote tmux | ❌ |
+
+Full honest head-to-head (our gaps included): [docs/COMPETITIVE.md](docs/COMPETITIVE.md).
+
+## The arsenal (40+ commands)
+
+- **`turtle-copilot`** — self-hosted AI co-pilot: error-explain, multi-turn chat, fix-memory recall, multi-backend (Claude / Ollama / Noetica)
+- **`turtle-autonomous`** — governed plan→execute→verify agent; safe-by-default dry-run; emits replayable, sealable reasoning evidence
+- **`turtle-drive`** — sovereign Warp Drive: parameterized workflows/runbooks, versioned in your own Gitea forge
+- **`turtle-share`** — live session sharing over WezTerm mux; unix-socket / SSH, no cloud account
+- **`turtle-gh` / `turtle-gitea`** — full `gh` CLI parity, Gitea-primary, with AI PR bodies + changelogs + review
+- **`turtle-seal`** — seal reasoning receipts into externally-verifiable, tamper-evident evidence
+- **`turtle-onboard`** guided first-run · **`turtle-diagnose`** health check · **`turtle-voice`** voice→shell · **`turtle-cost`** per-model AI spend
+- MCP server (`turtle-mcp-server`) exposes the whole fabric to Claude Code.
 
 TurtleTerm presents its own product surface: TurtleTerm launchers, TurtleTerm profile, TurtleTerm agent gateway, TurtleTerm release artifacts, and TurtleTerm skill manifests. Third-party notices and required license attribution are preserved in the repository license files and release artifacts.