Add HTTP gateway with streaming chat and multi-client conversation mapping (#1)

Reviewed-on: MoA/agent#1
Co-authored-by: highperfocused <highperfocused@pm.me>
Co-committed-by: highperfocused <highperfocused@pm.me>

.env.example

@@ -1,4 +1,19 @@
-# ── Ollama (local / self-hosted) ────────────────────────────────────────────────
+# ── Run mode ────────────────────────────────────────────────────────────────
+# gateway (default) starts HTTP API + streaming endpoint + simple web UI
+# single runs one-shot mode (PROMPT/stdin) and exits
+RUN_MODE=gateway
+
+# ── Gateway server ───────────────────────────────────────────────────────────
+GATEWAY_HOST=0.0.0.0
+GATEWAY_PORT=8787
+# Set to * or a specific origin for browser apps hosted elsewhere
+# GATEWAY_CORS_ORIGIN=*
+# Optional bearer token required on every API request
+# GATEWAY_AUTH_TOKEN=replace-me
+# Disable built-in web UI at /
+GATEWAY_ENABLE_WEB_UI=true
+
+# ── Ollama (local / self-hosted) ────────────────────────────────────────────
 # To use a local Ollama instance, override PROVIDER and MODEL:
 #   PROVIDER=ollama
 #   MODEL=llama3.2          # or qwen2.5-coder, mistral, deepseek-r1:14b …
@@ -12,7 +27,7 @@
 # Locally (no Docker): change to http://localhost:11434/v1
 #
 # OLLAMA_BASE_URL=http://host.docker.internal:11434/v1
-# OLLAMA_CONTEXT_WINDOW=32768  # tokens passed as num_ctx
+# OLLAMA_CONTEXT_WINDOW=32768
 # OLLAMA_MAX_TOKENS=8192
 
 # ── Model ────────────────────────────────────────────────────────────────────
@@ -22,7 +37,7 @@
 PROVIDER=anthropic
 MODEL=claude-sonnet-4-20250514
 
-# ── Authentication ────────────────────────────────────────────────────────────
+# ── Authentication ───────────────────────────────────────────────────────────
 # Generic API key for the selected PROVIDER (injected at runtime, not stored).
 API_KEY=sk-ant-...
 
@@ -36,36 +51,37 @@ API_KEY=sk-ant-...
 # XAI_API_KEY=...
 # OPENROUTER_API_KEY=...
 
-# ── Prompt ────────────────────────────────────────────────────────────────────
-# The user message to send. Alternatively pipe via stdin.
-PROMPT=List all .ts files in the current directory.
+# ── Prompt (single mode only) ───────────────────────────────────────────────
+# Used only when RUN_MODE=single.
+# PROMPT=List all .ts files in the current directory.
 
-# ── System prompt ─────────────────────────────────────────────────────────────
+# ── System prompt ────────────────────────────────────────────────────────────
 # Completely replaces the default system prompt (optional).
 # SYSTEM_PROMPT=You are a helpful assistant. Work inside /app only.
 
 # Appended to the (possibly overridden) system prompt (optional).
 # APPEND_SYSTEM_PROMPT=Always answer in English.
 
-# ── Thinking ──────────────────────────────────────────────────────────────────
-# off | minimal | low | medium | high | xhigh  (default: off)
+# ── Thinking ─────────────────────────────────────────────────────────────────
+# off | minimal | low | medium | high | xhigh
 THINKING_LEVEL=off
 
-# ── Tools ─────────────────────────────────────────────────────────────────────
-# all      → read, bash, edit, write  (default)
-# readonly → read only
+# ── Tools ────────────────────────────────────────────────────────────────────
+# all      → read, bash, edit, write (default)
+# readonly → read, grep, find, ls
 # none     → no built-in tools
-# Or a comma-separated list: read,bash
+# Or a comma-separated subset of coding tools: read,bash,edit,write
 TOOLS=all
 
-# ── Working directory ─────────────────────────────────────────────────────────
+# ── Working directory ────────────────────────────────────────────────────────
 # Directory the agent reads/writes. Maps to the Docker volume mount point.
 CWD=/app
 
-# ── Session persistence ───────────────────────────────────────────────────────
-# true → persist session files under CWD; false/unset → in-memory (default)
-SESSION_PERSIST=false
+# ── Session persistence ──────────────────────────────────────────────────────
+# true → persist agent sessions and gateway conversation index under CWD/.gateway
+# false → in-memory sessions only
+SESSION_PERSIST=true
 
-# ── Verbose tool logging ──────────────────────────────────────────────────────
+# ── Verbose tool logging ─────────────────────────────────────────────────────
 # true → log tool start/end events to stderr
 VERBOSE_TOOLS=false