llm-switcher 1.1.11 → 1.2.2

package/docs/cross-platform.md

@@ -110,18 +110,27 @@ no `hosts` file change and no certificate in a system trust store. See
 Build the certificates once. On Windows, run this line in Git Bash, not in PowerShell:
 
 ```bash
-bash blindfold/make-certs.sh chatgpt.com
+bash blindfold/make-certs.sh
 ```
 
-The subject of the certificate must match the host your Codex account calls:
+The leaf names exactly three hosts, and the interceptor routes exactly those three:
 
-| Codex sign-in | Host | Profile keys |
+| CONNECT host | Paths to this switcher's gateway | Everything else |
 | --- | --- | --- |
-| ChatGPT account | `chatgpt.com` | The defaults. Add nothing. |
-| API key | `api.openai.com` | `"blindfoldHost": "api.openai.com"`, `"blindfoldPrefix": "/v1"` |
+| `api.anthropic.com` | `/v1/messages`, `/v1/messages/...` | to `api.anthropic.com`, unchanged |
+| `api.openai.com` | `/v1/responses`, `/v1/responses/...`, `/v1/models`, `/v1/models/...` | to `api.openai.com`, unchanged |
+| `chatgpt.com` | `/backend-api/codex/...`, forwarded as `/v1` | to `chatgpt.com`, unchanged |
 
-`switch` compares the host in the profile against the subject alternative names of the
-leaf certificate. If they differ it refuses the activation, names the host, and prints
+A CONNECT host outside the table is tunneled untouched, with no TLS termination. A request
+whose `Host` header names a different host than its CONNECT target gets
+`421 Misdirected Request` and opens no upstream connection; the comparison ignores case and
+a default `:443`.
+
+There is no `blindfoldHost` and no `blindfoldPrefix` any more: this table is the routing, and
+it is not a profile setting.
+
+`switch` compares those three hosts against the subject alternative names of the leaf
+certificate. If one is missing it refuses the activation, names the host, and prints
 the command that rebuilds the certificate. No file is written on that path.
 
 ## Check that the platform work succeeded

package/docs/diagrams/ir-healer-pipeline.mmd

@@ -0,0 +1,16 @@
+sequenceDiagram
+    autonumber
+    actor CLI as Client Agent (Claude Code / Codex)
+    participant GW as LLM Switcher (:3456)
+    participant IR as Healer & IR Engine
+    participant UP as Upstream (intact / 9Router / Vertex)
+
+    CLI->>GW: POST /v1/messages or /v1/responses
+    GW->>IR: parseToIR(clientFormat, payload)
+    Note over IR: Repair orphan results, thinking budget & 1M context
+    IR-->>GW: emitUpstreamBody(outFormat, healedIR)
+    GW->>UP: POST upstream API (with AbortSignal)
+    UP-->>GW: Raw stream chunks (OpenAI / Vertex / Anthropic)
+    GW->>IR: normalizeUpstream(chunk) & parse reasoning
+    IR-->>GW: Native client event stream chunks
+    GW->>CLI: Synthesized genuine SSE (thinking_delta + text)

package/docs/diagrams/ir-healer-pipeline.svg

@@ -0,0 +1,90 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 949.355 455" width="949.355" height="455" style="--bg:#1a1b26;--fg:#a9b1d6;--line:#3d59a1;--accent:#7aa2f7;--muted:#565f89;background:var(--bg)">
+<style>
+  @import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&amp;display=swap');
+  text { font-family: 'Inter', system-ui, sans-serif; }
+  svg {
+    /* Derived from --bg and --fg (overridable via --line, --accent, etc.) */
+    --_text:          var(--fg);
+    --_text-sec:      var(--muted, color-mix(in srgb, var(--fg) 60%, var(--bg)));
+    --_text-muted:    var(--muted, color-mix(in srgb, var(--fg) 40%, var(--bg)));
+    --_text-faint:    color-mix(in srgb, var(--fg) 25%, var(--bg));
+    --_line:          var(--line, color-mix(in srgb, var(--fg) 50%, var(--bg)));
+    --_arrow:         var(--accent, color-mix(in srgb, var(--fg) 85%, var(--bg)));
+    --_node-fill:     var(--surface, color-mix(in srgb, var(--fg) 3%, var(--bg)));
+    --_node-stroke:   var(--border, color-mix(in srgb, var(--fg) 20%, var(--bg)));
+    --_group-fill:    var(--bg);
+    --_group-hdr:     color-mix(in srgb, var(--fg) 5%, var(--bg));
+    --_inner-stroke:  color-mix(in srgb, var(--fg) 12%, var(--bg));
+    --_key-badge:     color-mix(in srgb, var(--fg) 10%, var(--bg));
+  }
+</style>
+<defs>
+  <marker id="seq-arrow" markerWidth="8" markerHeight="5" refX="8" refY="2.5" orient="auto-start-reverse">
+    <polygon points="0 0, 8 2.5, 0 5" fill="var(--_arrow)" />
+  </marker>
+  <marker id="seq-arrow-open" markerWidth="8" markerHeight="5" refX="8" refY="2.5" orient="auto-start-reverse">
+    <polyline points="0 0, 8 2.5, 0 5" fill="none" stroke="var(--_arrow)" stroke-width="1" />
+  </marker>
+</defs>
+<line class="lifeline" data-actor="CLI" x1="147.01" y1="70" x2="147.01" y2="425" stroke="var(--_line)" stroke-width="0.75" stroke-dasharray="6 4" />
+<line class="lifeline" data-actor="GW" x1="383.6095" y1="70" x2="383.6095" y2="425" stroke="var(--_line)" stroke-width="0.75" stroke-dasharray="6 4" />
+<line class="lifeline" data-actor="IR" x1="573.8965000000001" y1="70" x2="573.8965000000001" y2="425" stroke="var(--_line)" stroke-width="0.75" stroke-dasharray="6 4" />
+<line class="lifeline" data-actor="UP" x1="801.9745" y1="70" x2="801.9745" y2="425" stroke="var(--_line)" stroke-width="0.75" stroke-dasharray="6 4" />
+<g class="message" data-from="CLI" data-to="GW" data-label="POST /v1/messages or /v1/responses" data-line-style="solid" data-arrow-head="filled" data-self="false">
+  <line x1="147.01" y1="90" x2="383.6095" y2="90" stroke="var(--_line)" stroke-width="1" marker-end="url(#seq-arrow)" />
+  <text x="265.30975" y="80" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">POST /v1/messages or /v1/responses</text>
+</g>
+<g class="message" data-from="GW" data-to="IR" data-label="parseToIR(clientFormat, payload)" data-line-style="solid" data-arrow-head="filled" data-self="false">
+  <line x1="383.6095" y1="130" x2="573.8965000000001" y2="130" stroke="var(--_line)" stroke-width="1" marker-end="url(#seq-arrow)" />
+  <text x="478.75300000000004" y="120" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">parseToIR(clientFormat, payload)</text>
+</g>
+<g class="message" data-from="IR" data-to="GW" data-label="emitUpstreamBody(outFormat, healedIR)" data-line-style="dashed" data-arrow-head="filled" data-self="false">
+  <line x1="573.8965000000001" y1="185" x2="383.6095" y2="185" stroke="var(--_line)" stroke-width="1" stroke-dasharray="6 4" marker-end="url(#seq-arrow)" />
+  <text x="478.75300000000004" y="175" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">emitUpstreamBody(outFormat, healedIR)</text>
+</g>
+<g class="message" data-from="GW" data-to="UP" data-label="POST upstream API (with AbortSignal)" data-line-style="solid" data-arrow-head="filled" data-self="false">
+  <line x1="383.6095" y1="225" x2="801.9745" y2="225" stroke="var(--_line)" stroke-width="1" marker-end="url(#seq-arrow)" />
+  <text x="592.792" y="215" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">POST upstream API (with AbortSignal)</text>
+</g>
+<g class="message" data-from="UP" data-to="GW" data-label="Raw stream chunks (OpenAI / Vertex / Anthropic)" data-line-style="dashed" data-arrow-head="filled" data-self="false">
+  <line x1="801.9745" y1="265" x2="383.6095" y2="265" stroke="var(--_line)" stroke-width="1" stroke-dasharray="6 4" marker-end="url(#seq-arrow)" />
+  <text x="592.792" y="255" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">Raw stream chunks (OpenAI / Vertex / Anthropic)</text>
+</g>
+<g class="message" data-from="GW" data-to="IR" data-label="normalizeUpstream(chunk) &amp; parse reasoning" data-line-style="solid" data-arrow-head="filled" data-self="false">
+  <line x1="383.6095" y1="305" x2="573.8965000000001" y2="305" stroke="var(--_line)" stroke-width="1" marker-end="url(#seq-arrow)" />
+  <text x="478.75300000000004" y="295" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">normalizeUpstream(chunk) &amp; parse reasoning</text>
+</g>
+<g class="message" data-from="IR" data-to="GW" data-label="Native client event stream chunks" data-line-style="dashed" data-arrow-head="filled" data-self="false">
+  <line x1="573.8965000000001" y1="345" x2="383.6095" y2="345" stroke="var(--_line)" stroke-width="1" stroke-dasharray="6 4" marker-end="url(#seq-arrow)" />
+  <text x="478.75300000000004" y="335" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">Native client event stream chunks</text>
+</g>
+<g class="message" data-from="GW" data-to="CLI" data-label="Synthesized genuine SSE (thinking_delta + text)" data-line-style="solid" data-arrow-head="filled" data-self="false">
+  <line x1="383.6095" y1="385" x2="147.01" y2="385" stroke="var(--_line)" stroke-width="1" marker-end="url(#seq-arrow)" />
+  <text x="265.30975" y="375" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">Synthesized genuine SSE (thinking_delta + text)</text>
+</g>
+<g class="note" data-position="over" data-actors="IR">
+  <polygon points="443.4595000000001,138 698.3335,138 704.3335,144 704.3335,161 443.4595000000001,161" fill="var(--bg)" stroke="var(--_node-stroke)" stroke-width="0.75" />
+  <polygon points="698.3335,138 704.3335,144 698.3335,144" fill="var(--_inner-stroke)" stroke="var(--_node-stroke)" stroke-width="0.75" />
+  <text x="573.8965000000001" y="149.5" font-size="11" text-anchor="middle" font-weight="400" fill="var(--_text-muted)" dy="3.8499999999999996">Repair orphan results, thinking budget &amp; 1M context</text>
+</g>
+<g class="actor" data-id="CLI" data-label="Client Agent (Claude Code / Codex)" data-type="actor">
+  <g transform="translate(129.01,32) scale(1.5)">
+    <path d="M21 12C21 16.9706 16.9706 21 12 21C7.02944 21 3 16.9706 3 12C3 7.02944 7.02944 3 12 3C16.9706 3 21 7.02944 21 12Z" fill="none" stroke="var(--_line)" stroke-width="0.6666666666666666" />
+    <path d="M15 10C15 11.6569 13.6569 13 12 13C10.3431 13 9 11.6569 9 10C9 8.34315 10.3431 7 12 7C13.6569 7 15 8.34315 15 10Z" fill="none" stroke="var(--_line)" stroke-width="0.6666666666666666" />
+    <path d="M5.62842 18.3563C7.08963 17.0398 9.39997 16 12 16C14.6 16 16.9104 17.0398 18.3716 18.3563" fill="none" stroke="var(--_line)" stroke-width="0.6666666666666666" />
+  </g>
+  <text x="147.01" y="84" font-size="13" text-anchor="middle" font-weight="500" fill="var(--_text)" dy="4.55">Client Agent (Claude Code / Codex)</text>
+</g>
+<g class="actor" data-id="GW" data-label="LLM Switcher (:3456)" data-type="participant">
+  <rect x="304.02" y="30" width="159.17900000000003" height="40" rx="4" ry="4" fill="var(--_node-fill)" stroke="var(--_node-stroke)" stroke-width="1" />
+  <text x="383.6095" y="50" font-size="13" text-anchor="middle" font-weight="500" fill="var(--_text)" dy="4.55">LLM Switcher (:3456)</text>
+</g>
+<g class="actor" data-id="IR" data-label="Healer &amp; IR Engine" data-type="participant">
+  <rect x="503.19900000000007" y="30" width="141.39499999999998" height="40" rx="4" ry="4" fill="var(--_node-fill)" stroke="var(--_node-stroke)" stroke-width="1" />
+  <text x="573.8965000000001" y="50" font-size="13" text-anchor="middle" font-weight="500" fill="var(--_text)" dy="4.55">Healer &amp; IR Engine</text>
+</g>
+<g class="actor" data-id="UP" data-label="Upstream (intact / 9Router / Vertex)" data-type="participant">
+  <rect x="684.594" y="30" width="234.76099999999997" height="40" rx="4" ry="4" fill="var(--_node-fill)" stroke="var(--_node-stroke)" stroke-width="1" />
+  <text x="801.9745" y="50" font-size="13" text-anchor="middle" font-weight="500" fill="var(--_text)" dy="4.55">Upstream (intact / 9Router / Vertex)</text>
+</g>
+</svg>