Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

fix(codex): configure recall compressor base URL (#3601)

Tequila Sunset9042a0254f92
diff --git a/examples/codex-memory-plugin/DESIGN.md b/examples/codex-memory-plugin/DESIGN.md
index dccd9c128..19254cb19 100644
--- a/examples/codex-memory-plugin/DESIGN.md
+++ b/examples/codex-memory-plugin/DESIGN.md
@@ -253,6 +253,7 @@ Env var overrides for tuning without rebuilding:
| `OPENVIKING_RECALL_COMPRESS` | `1` | set `0` / `off` to skip `codex exec` compression |
| `OPENVIKING_RECALL_COMPRESS_MODEL` | unset | custom first-choice compressor model; `off` disables compression |
| `OPENVIKING_RECALL_COMPRESS_THINKING` | unset | custom `model_reasoning_effort`; `default` means omit override; alias `OPENVIKING_RECALL_COMPRESS_REASONING_EFFORT` |
+| `OPENVIKING_RECALL_COMPRESS_BASE_URL` | unset | custom API base URL for the nested `codex exec` compressor |
| `OPENVIKING_RECALL_COMPRESS_DETECT_ON_STARTUP` | `1` | recreate/cache compressor profile during every `SessionStart` |
| `OPENVIKING_RECALL_COMPRESS_DETECT_TIMEOUT_MS` | `15000` | per-candidate compressor probe timeout |
| `OPENVIKING_RECALL_COMPRESS_DETECT_TTL_MS` | `604800000` (7 days) | cache TTL used by `UserPromptSubmit` reads |
@@ -285,6 +286,20 @@ both:
codex -m <model> -c 'model_reasoning_effort="low"' exec ...
```
+The compressor runs with `--ignore-user-config`, so it does not inherit the
+main Codex process's provider table. When
+`OPENVIKING_RECALL_COMPRESS_BASE_URL` is set, the plugin adds an isolated
+provider for the nested request:
+
+```bash
+-c 'model_provider="openviking_compressor"' \
+-c 'model_providers.openviking_compressor.name="openviking_compressor"' \
+-c 'model_providers.openviking_compressor.base_url="<url>"'
+```
+
+The selected model and thinking effort remain profile data; the base URL is
+runtime configuration and is supplied when the command is built.
+
`thinking=default` omits the `model_reasoning_effort` override. This is
important for model families whose default effort is tuned by Codex.
diff --git a/examples/codex-memory-plugin/README.md b/examples/codex-memory-plugin/README.md
index e180652c5..849eab33d 100644
--- a/examples/codex-memory-plugin/README.md
+++ b/examples/codex-memory-plugin/README.md
@@ -129,6 +129,7 @@ export OPENVIKING_RECALL_LIMIT=10
export OPENVIKING_RECALL_COMPRESS=1
export OPENVIKING_RECALL_COMPRESS_MODEL=gpt-5.3-codex-spark
export OPENVIKING_RECALL_COMPRESS_THINKING=default
+export OPENVIKING_RECALL_COMPRESS_BASE_URL=https://api.example.com/v1
export OPENVIKING_RECALL_TIMEOUT_MS=120000
export OPENVIKING_CAPTURE_ASSISTANT_TURNS=1
export OPENVIKING_AUTO_COMMIT_ON_COMPACT=1
@@ -226,6 +227,7 @@ Config knobs:
| `OPENVIKING_RECALL_COMPRESS` | `1` | Set `0` / `off` to disable `codex exec` compression. |
| `OPENVIKING_RECALL_COMPRESS_MODEL` | unset | Custom first-choice compressor model. Set `off` to disable compression. |
| `OPENVIKING_RECALL_COMPRESS_THINKING` | unset | Custom `model_reasoning_effort`; `default` omits the Codex config override. Alias: `OPENVIKING_RECALL_COMPRESS_REASONING_EFFORT`. |
+| `OPENVIKING_RECALL_COMPRESS_BASE_URL` | unset | Base URL for the nested compressor's provider. Use this when `--ignore-user-config` prevents the compressor from reading the main Codex provider configuration. |
| `OPENVIKING_RECALL_COMPRESS_DETECT_ON_STARTUP` | `1` | Recreate/cache compressor profile in `SessionStart`. |
| `OPENVIKING_RECALL_COMPRESS_DETECT_TIMEOUT_MS` | `15000` | Per-candidate startup probe timeout. |
| `OPENVIKING_RECALL_COMPRESS_DETECT_TTL_MS` | `604800000` | Cache TTL used by `UserPromptSubmit` when reading the latest profile. |
diff --git a/examples/codex-memory-plugin/scripts/auto-recall.mjs b/examples/codex-memory-plugin/scripts/auto-recall.mjs
index 8489e6124..2fd0d9360 100644
--- a/examples/codex-memory-plugin/scripts/auto-recall.mjs
+++ b/examples/codex-memory-plugin/scripts/auto-recall.mjs
@@ -421,7 +421,7 @@ async function getRecallCompressorProfile() {
async function runCodexCompressor(prompt, profile) {
const tmp = await mkdtemp(join(tmpdir(), "ov-recall-compress-"));
const outputPath = join(tmp, "last-message.txt");
- const args = buildCodexExecArgs(profile, outputPath);
+ const args = buildCodexExecArgs(profile, outputPath, cfg);
try {
return await new Promise((resolve) => {
diff --git a/examples/codex-memory-plugin/scripts/auto-recall.test.mjs b/examples/codex-memory-plugin/scripts/auto-recall.test.mjs
index 1242c277d..1273b5f44 100644
--- a/examples/codex-memory-plugin/scripts/auto-recall.test.mjs
+++ b/examples/codex-memory-plugin/scripts/auto-recall.test.mjs
@@ -117,6 +117,7 @@ async function withFakeCodex(output, fn, { exitCode = 0 } = {}) {
"codex.js",
);
const callLog = join(binDir, "calls.log");
+ const argsLog = join(binDir, "args.log");
const fakeCodex = `#!/usr/bin/env node
const fs = require("node:fs");
@@ -124,6 +125,7 @@ const outputFlagIndex = process.argv.indexOf("--output-last-message");
const outputPath = outputFlagIndex >= 0 ? process.argv[outputFlagIndex + 1] : "";
fs.readFileSync(0);
fs.appendFileSync(process.env.FAKE_CODEX_CALL_LOG, "called\\n");
+fs.appendFileSync(process.env.FAKE_CODEX_ARGS_LOG, process.argv.slice(2).join("\\n") + "\\n");
const exitCode = Number(process.env.FAKE_CODEX_EXIT_CODE || 0);
if (exitCode !== 0) process.exit(exitCode);
@@ -137,9 +139,11 @@ fs.writeFileSync(outputPath, process.env.FAKE_CODEX_OUTPUT || "");
try {
return await fn({
callLog,
+ argsLog,
env: {
PATH: `${binDir}${delimiter}${process.env.PATH}`,
FAKE_CODEX_CALL_LOG: callLog,
+ FAKE_CODEX_ARGS_LOG: argsLog,
FAKE_CODEX_EXIT_CODE: String(exitCode),
FAKE_CODEX_OUTPUT: output,
},
@@ -160,7 +164,7 @@ async function runEndpointCompressionCase({
const stateDir = await mkdtemp(join(tmpdir(), "ov-auto-recall-endpoint-compress-"));
let requestBody = null;
try {
- return await withFakeCodex(compressorOutput, async ({ callLog, env }) => {
+ return await withFakeCodex(compressorOutput, async ({ callLog, argsLog, env }) => {
const result = await withMockOpenViking(async (req, res) => {
const url = new URL(req.url, "http://127.0.0.1");
if (req.method === "GET" && url.pathname === "/health") {
@@ -197,9 +201,11 @@ async function runEndpointCompressionCase({
},
));
const compressorCallLog = await readFile(callLog, "utf-8").catch(() => "");
+ const compressorArgs = await readFile(argsLog, "utf-8").catch(() => "");
return {
output: JSON.parse(result.stdout.trim()),
compressorCalls: compressorCallLog.trim().split("\n").filter(Boolean).length,
+ compressorArgs: compressorArgs.trim().split("\n").filter(Boolean),
requestBody,
};
}, { exitCode });
@@ -382,6 +388,26 @@ test("auto-recall applies the relevance compressor to server recall entries", as
assert.equal(result.requestBody.max_chars, 18000);
});
+test("auto-recall passes the configured compressor base URL to Codex", async () => {
+ const result = await runEndpointCompressionCase({
+ prompt: "Explain HTTP 429",
+ entry: {
+ uri: "viking://user/zeus/memories/events/retry.md",
+ score: 0.91,
+ type: "events",
+ mode: "summary",
+ summary: "Retry with backoff",
+ },
+ rendered: "<memory_group>Retry with backoff</memory_group>",
+ compressorOutput: "NO_RELEVANT_MEMORY",
+ extraEnv: { OPENVIKING_RECALL_COMPRESS_BASE_URL: "https://compressor.example/v1" },
+ });
+
+ assert.ok(result.compressorArgs.includes('model_provider="openviking_compressor"'));
+ assert.ok(result.compressorArgs.includes('model_providers.openviking_compressor.name="openviking_compressor"'));
+ assert.ok(result.compressorArgs.includes('model_providers.openviking_compressor.base_url="https://compressor.example/v1"'));
+});
+
test("auto-recall falls back to a bounded deterministic digest when endpoint compression fails", async () => {
const result = await runEndpointCompressionCase({
prompt: "Which editor do I prefer?",
diff --git a/examples/codex-memory-plugin/scripts/config.mjs b/examples/codex-memory-plugin/scripts/config.mjs
index b8e393d57..477416b41 100644
--- a/examples/codex-memory-plugin/scripts/config.mjs
+++ b/examples/codex-memory-plugin/scripts/config.mjs
@@ -34,6 +34,7 @@
* OPENVIKING_TIMEOUT_MS, OPENVIKING_CAPTURE_TIMEOUT_MS
* OPENVIKING_RECALL_TIMEOUT_MS, OPENVIKING_RECALL_COMPRESS_TIMEOUT_MS
* OPENVIKING_RECALL_COMPRESS_MODEL, OPENVIKING_RECALL_COMPRESS_THINKING
+ * OPENVIKING_RECALL_COMPRESS_BASE_URL
* OPENVIKING_RECALL_LIMIT, OPENVIKING_SCORE_THRESHOLD
* OPENVIKING_WORKSPACE_PEER, OPENVIKING_RECALL_PEER_SCOPE
* OPENVIKING_NO_AUTO_INJECT, OPENVIKING_PROFILE_TOKEN_BUDGET
@@ -139,6 +140,10 @@ export function loadConfig() {
process.env.OPENVIKING_RECALL_COMPRESS_MODEL,
hasOwn(cx, "recallCompressModel") ? str(cx.recallCompressModel, "") : "",
);
+ const recallCompressBaseUrl = str(
+ process.env.OPENVIKING_RECALL_COMPRESS_BASE_URL,
+ hasOwn(cx, "recallCompressBaseUrl") ? str(cx.recallCompressBaseUrl, "") : "",
+ );
const cxRecallCompressThinking = hasOwn(cx, "recallCompressThinking")
? cx.recallCompressThinking
: (hasOwn(cx, "recallCompressReasoningEffort") ? cx.recallCompressReasoningEffort : "");
@@ -186,6 +191,7 @@ export function loadConfig() {
recallPeerScope,
recallCompress: envBool("OPENVIKING_RECALL_COMPRESS") ?? configBool(cx.recallCompress, true),
recallCompressModel,
+ recallCompressBaseUrl,
recallCompressThinking,
recallCompressConfigured: Boolean(recallCompressModel || recallCompressThinking),
recallCompressTimeoutMs,
diff --git a/examples/codex-memory-plugin/scripts/recall-compressor-profile.mjs b/examples/codex-memory-plugin/scripts/recall-compressor-profile.mjs
index 5e7b32e8f..21e1f0ea6 100644
--- a/examples/codex-memory-plugin/scripts/recall-compressor-profile.mjs
+++ b/examples/codex-memory-plugin/scripts/recall-compressor-profile.mjs
@@ -8,6 +8,7 @@ const DEFAULT_PRIMARY = { model: "gpt-5.3-codex-spark", thinking: "default", sou
const DEFAULT_FALLBACK = { model: "gpt-5.6-luna", thinking: "low", source: "default_fallback" };
const PROFILE_SCHEMA_VERSION = 3;
const DEFAULT_CODEX_HOME = join(homedir(), ".codex");
+const COMPRESSOR_PROVIDER = "openviking_compressor";
function isOff(value) {
return /^(?:0|false|no|off|none|disabled)$/i.test(String(value || "").trim());
@@ -27,12 +28,19 @@ export function recallCompressionExplicitlyOff(cfg) {
return !cfg.recallCompress || isOff(cfg.recallCompressModel) || isOff(cfg.recallCompressThinking);
}
-export function buildCodexExecArgs(profile, outputPath) {
+export function buildCodexExecArgs(profile, outputPath, cfg = {}) {
const args = [];
if (profile.model) args.push("-m", profile.model);
if (profile.thinking && profile.thinking !== "default") {
args.push("-c", `model_reasoning_effort=${JSON.stringify(profile.thinking)}`);
}
+ if (cfg.recallCompressBaseUrl) {
+ args.push(
+ "-c", `model_provider=${JSON.stringify(COMPRESSOR_PROVIDER)}`,
+ "-c", `model_providers.${COMPRESSOR_PROVIDER}.name=${JSON.stringify(COMPRESSOR_PROVIDER)}`,
+ "-c", `model_providers.${COMPRESSOR_PROVIDER}.base_url=${JSON.stringify(cfg.recallCompressBaseUrl)}`,
+ );
+ }
args.push(
"--sandbox",
"read-only",
diff --git a/examples/codex-memory-plugin/scripts/recall-compressor-profile.test.mjs b/examples/codex-memory-plugin/scripts/recall-compressor-profile.test.mjs
index 2ea21a1f3..2101a0112 100644
--- a/examples/codex-memory-plugin/scripts/recall-compressor-profile.test.mjs
+++ b/examples/codex-memory-plugin/scripts/recall-compressor-profile.test.mjs
@@ -5,6 +5,7 @@ import { join } from "node:path";
import test from "node:test";
import {
+ buildCodexExecArgs,
detectRecallCompressorProfile,
invalidateRecallCompressorProfileCache,
loadCachedRecallCompressorProfile,
@@ -50,6 +51,31 @@ async function writeModelsCache(codexHome, slugs) {
);
}
+test("buildCodexExecArgs injects a configured compressor base URL", () => {
+ const args = buildCodexExecArgs(
+ { model: "gpt-5.3-codex-spark", thinking: "default" },
+ "/tmp/last-message.txt",
+ { recallCompressBaseUrl: "https://compressor.example/v1" },
+ );
+
+ assert.deepEqual(args.slice(0, 8), [
+ "-m", "gpt-5.3-codex-spark",
+ "-c", 'model_provider="openviking_compressor"',
+ "-c", 'model_providers.openviking_compressor.name="openviking_compressor"',
+ "-c", 'model_providers.openviking_compressor.base_url="https://compressor.example/v1"',
+ ]);
+ assert.ok(args.indexOf("exec") > args.indexOf('model_providers.openviking_compressor.base_url="https://compressor.example/v1"'));
+});
+
+test("buildCodexExecArgs preserves the existing command without a compressor base URL", () => {
+ const args = buildCodexExecArgs(
+ { model: "gpt-5.3-codex-spark", thinking: "default" },
+ "/tmp/last-message.txt",
+ );
+
+ assert.equal(args.some((arg) => arg.includes("openviking_compressor")), false);
+});
+
test("loadCodexModelsCache returns empty set when cache missing", async () => {
await withTempState(async ({ codexHome }) => {
const result = await loadCodexModelsCache({ CODEX_HOME: codexHome });