diff --git a/hypaware-core/plugins-workspace/ai-gateway/src/recorder.js b/hypaware-core/plugins-workspace/ai-gateway/src/recorder.js
index 0d19d62..3e49055 100644
--- a/hypaware-core/plugins-workspace/ai-gateway/src/recorder.js
+++ b/hypaware-core/plugins-workspace/ai-gateway/src/recorder.js
@@ -1,6 +1,7 @@
 // @ts-check
 
 import { randomBytes } from 'node:crypto'
+import { brotliDecompressSync, gunzipSync, inflateRawSync, inflateSync } from 'node:zlib'
 
 import { isSseHeaders, SseParser } from './sse.js'
 
@@ -116,6 +117,12 @@ export class Exchange {
     this.requestBytes = 0
     /** @type {ResponseStart | undefined} */
     this.response = undefined
+    /**
+     * Raw upstream `content-encoding`, captured before header redaction so
+     * body decoding still works when an operator redacts that header.
+     * @type {string | string[] | undefined}
+     */
+    this.responseContentEncoding = undefined
     /** @type {Buffer[]} */
     this.responseChunks = []
     /** @type {number} */
@@ -178,6 +185,7 @@ export class Exchange {
       status: init.status,
       headers: redactHeaders(init.headers, this.redactSet),
     }
+    this.responseContentEncoding = headerValue(init.headers, 'content-encoding')
     this.isSse = isSseHeaders(init.headers)
   }
 
@@ -256,10 +264,17 @@ export class Exchange {
     this.finished = true
 
     const tsEndMs = Date.now()
-    const requestBody = Buffer.concat(this.requestChunks).toString('utf8')
+    // The proxy is a pass-through, so a body carries whatever
+    // `content-encoding` the upstream (or client) applied. Decode it
+    // before stringifying or a gzip/br/deflate body lands in the cache
+    // as mojibake that no downstream projector can parse as JSON.
+    const requestBody = decodeBody(
+      Buffer.concat(this.requestChunks),
+      headerValue(this._rawRequestHeaders, 'content-encoding')
+    )
     const responseBody = this.isSse
       ? null
-      : Buffer.concat(this.responseChunks).toString('utf8')
+      : decodeBody(Buffer.concat(this.responseChunks), this.responseContentEncoding)
     const devRunId = this.devRunIdFromHeaders()
     /** @type {Record<string, unknown>} */
     const metadata = {}
@@ -294,6 +309,68 @@ export class Exchange {
   }
 }
 
+/**
+ * Reverse the `content-encoding` applied to a captured body and decode
+ * it as UTF-8. Best-effort: an empty buffer yields `''`, an unknown or
+ * undecodable encoding falls back to the raw bytes so a (possibly
+ * garbled) row is still written rather than the exchange being dropped.
+ *
+ * @param {Buffer} buf
+ * @param {string | string[] | undefined} encodingHeader
+ * @returns {string}
+ */
+function decodeBody(buf, encodingHeader) {
+  if (buf.byteLength === 0) return ''
+  const encodings = parseEncodings(encodingHeader)
+  if (encodings.length === 0) return buf.toString('utf8')
+  let current = buf
+  // `content-encoding` lists transforms in the order they were applied;
+  // decode in reverse to undo them.
+  for (let i = encodings.length - 1; i >= 0; i--) {
+    const enc = encodings[i]
+    try {
+      if (enc === 'gzip' || enc === 'x-gzip') current = gunzipSync(current)
+      else if (enc === 'br') current = brotliDecompressSync(current)
+      else if (enc === 'deflate') current = inflateOrRaw(current)
+      else return current.toString('utf8') // unknown codec — stop, keep what we have
+    } catch {
+      return buf.toString('utf8') // undecodable — fall back to the raw bytes
+    }
+  }
+  return current.toString('utf8')
+}
+
+/**
+ * Some servers emit raw DEFLATE without the zlib header. Try the
+ * conformant decoder first, then the headerless variant.
+ *
+ * @param {Buffer} buf
+ * @returns {Buffer}
+ */
+function inflateOrRaw(buf) {
+  try {
+    return inflateSync(buf)
+  } catch {
+    return inflateRawSync(buf)
+  }
+}
+
+/**
+ * Normalize a `content-encoding` header into an ordered list of lowercase
+ * codec tokens, dropping `identity` and empties.
+ *
+ * @param {string | string[] | undefined} header
+ * @returns {string[]}
+ */
+function parseEncodings(header) {
+  if (header === undefined) return []
+  const raw = Array.isArray(header) ? header.join(',') : header
+  return raw
+    .split(',')
+    .map((token) => token.trim().toLowerCase())
+    .filter((token) => token.length > 0 && token !== 'identity')
+}
+
 /**
  * @param {readonly string[] | undefined} extra
  */
diff --git a/test/plugins/ai-gateway-recorder-encoding.test.js b/test/plugins/ai-gateway-recorder-encoding.test.js
new file mode 100644
index 0000000..0cbf964
--- /dev/null
+++ b/test/plugins/ai-gateway-recorder-encoding.test.js
@@ -0,0 +1,115 @@
+// @ts-check
+
+import assert from 'node:assert/strict'
+import test from 'node:test'
+import { brotliCompressSync, deflateRawSync, deflateSync, gzipSync } from 'node:zlib'
+
+import { createRecorder } from '../../hypaware-core/plugins-workspace/ai-gateway/src/recorder.js'
+
+/**
+ * @param {{ responseHeaders?: Record<string, string | string[]>, responseBody?: Buffer, requestHeaders?: Record<string, string>, requestBody?: Buffer, redactHeaders?: string[] }} opts
+ */
+function finishExchange(opts) {
+  const recorder = createRecorder({ redactHeaders: opts.redactHeaders })
+  const exchange = recorder.startExchange({
+    upstream: 'test',
+    provider: 'test',
+    method: 'POST',
+    path: '/v1/messages',
+    requestHeaders: opts.requestHeaders ?? {},
+  })
+  if (opts.requestBody) exchange.appendRequestChunk(opts.requestBody)
+  exchange.setResponseStart({ status: 200, headers: opts.responseHeaders ?? {} })
+  if (opts.responseBody) exchange.appendResponseChunk(opts.responseBody)
+  return exchange.finalize()
+}
+
+const payload = JSON.stringify({ id: 'msg_1', content: [{ type: 'text', text: 'hello' }] })
+
+test('decodes a gzip-encoded response body', () => {
+  const row = finishExchange({
+    responseHeaders: { 'content-encoding': 'gzip' },
+    responseBody: gzipSync(Buffer.from(payload)),
+  })
+  assert.equal(row.response_body, payload)
+})
+
+test('decodes a brotli-encoded response body', () => {
+  const row = finishExchange({
+    responseHeaders: { 'content-encoding': 'br' },
+    responseBody: brotliCompressSync(Buffer.from(payload)),
+  })
+  assert.equal(row.response_body, payload)
+})
+
+test('decodes a deflate-encoded response body (zlib and raw)', () => {
+  const zlib = finishExchange({
+    responseHeaders: { 'content-encoding': 'deflate' },
+    responseBody: deflateSync(Buffer.from(payload)),
+  })
+  assert.equal(zlib.response_body, payload)
+
+  const raw = finishExchange({
+    responseHeaders: { 'content-encoding': 'deflate' },
+    responseBody: deflateRawSync(Buffer.from(payload)),
+  })
+  assert.equal(raw.response_body, payload)
+})
+
+test('passes through an identity / unencoded response body unchanged', () => {
+  const identity = finishExchange({
+    responseHeaders: { 'content-encoding': 'identity' },
+    responseBody: Buffer.from(payload),
+  })
+  assert.equal(identity.response_body, payload)
+
+  const none = finishExchange({ responseBody: Buffer.from(payload) })
+  assert.equal(none.response_body, payload)
+})
+
+test('decodes a gzip-encoded request body', () => {
+  const row = finishExchange({
+    requestHeaders: { 'content-encoding': 'gzip' },
+    requestBody: gzipSync(Buffer.from(payload)),
+  })
+  assert.equal(row.request_body, payload)
+})
+
+test('handles content-encoding header case-insensitively', () => {
+  const row = finishExchange({
+    responseHeaders: { 'Content-Encoding': 'GZIP' },
+    responseBody: gzipSync(Buffer.from(payload)),
+  })
+  assert.equal(row.response_body, payload)
+})
+
+test('decodes the body even when content-encoding is a redacted header', () => {
+  // The raw upstream encoding is captured before redaction, so decoding
+  // still works when an operator adds content-encoding to redactHeaders.
+  const row = finishExchange({
+    redactHeaders: ['content-encoding'],
+    responseHeaders: { 'content-encoding': 'gzip' },
+    responseBody: gzipSync(Buffer.from(payload)),
+  })
+  assert.equal(row.response_body, payload)
+  // The stored header is still redacted in the row.
+  const headers = JSON.parse(/** @type {string} */ (row.response_headers))
+  assert.match(headers['content-encoding'], /^REDACTED:/)
+})
+
+test('falls back to raw bytes when the encoding is unknown', () => {
+  const row = finishExchange({
+    responseHeaders: { 'content-encoding': 'snappy' },
+    responseBody: Buffer.from(payload),
+  })
+  assert.equal(row.response_body, payload)
+})
+
+test('falls back to raw bytes when a gzip body is corrupt', () => {
+  const corrupt = Buffer.from('not actually gzip')
+  const row = finishExchange({
+    responseHeaders: { 'content-encoding': 'gzip' },
+    responseBody: corrupt,
+  })
+  assert.equal(row.response_body, corrupt.toString('utf8'))
+})