evals-lab 0.10.0 → 0.11.0

package/CHANGELOG.md

@@ -5,6 +5,44 @@ Newest first. Read a release's **Upgrade notes** before installing it: an
 upgrade can rewrite what the lab keeps in your data directory, and an older
 version cannot always read it back.
 
+## 0.11.0
+
+### Upgrade notes
+
+- **Copy your data directory before upgrading** (see "User data" in the
+  README). A pipeline 0.11.0 saves is pipeline version 16, which 0.10.x
+  cannot read. To go back, reinstall 0.10.0 and restore the copy.
+- On first start, each Target profile becomes an AI Connection (its type,
+  address, key and model), keeping its key and who it is shared with, plus a
+  Connection Profile where it changed any model or image setting. The old
+  profiles stay stored, unused.
+- A pipeline's target names a Connection and a Profile. Pipelines read as
+  version 16, and are saved as it at their next edit; a run from before
+  shows its Target profile as its Profile.
+
+### Added
+
+- Setup › Connections holds AI Connections and Connection Profiles: a
+  connection is where to send and with which key, and a profile is how to
+  ask. A target picks one of each, so one connection serves several
+  profiles.
+- Every pipeline, prompt, run, profile and wizard has its own address, and
+  Copy link copies it.
+- A run's Responses read as a flow, one ordered table a stage, and name the
+  stage and filter that discarded a reply. Move to filter set… and Move all
+  turn a filter into part of a filter set.
+- A Call is viewed, edited, renamed and cloned in one dialog, and a
+  workflow or dynamic Source's settings are edited where they were made.
+- The package page gives a contact address and links to its home page.
+
+### Fixed
+
+- File on disk never reads `/proc`, and a Call holding the lab's own secret
+  is refused.
+- Importing a version-15 YAML pipeline keeps each target's profile.
+- A Target profile from before reply limits keeps its 160-token cap when it
+  becomes a connection.
+
 ## 0.10.0
 
 ### Upgrade notes
@@ -42,7 +80,9 @@ version cannot always read it back.
     lab warns at start when one is on while it listens beyond this machine
     without `LAB_PASSWORD`. `SOURCE_LOCAL_ROOT=<dir>` lets File on disk read only
     under that directory. A command runs without a shell, in an empty
-    directory, without the lab's own secrets.
+    directory, with the lab's own secrets out of its environment -- but it
+    can still read them from the lab's, so turn Command on only where
+    everyone who can reach the lab may see them.
 - Copy or move… copies or moves a pipeline, dataset, filter set, prompt or
   Source to another workspace, and says first what a move would leave broken.
 - A run's view opens on an Evals summary, and Show details shows what came

package/README.md

@@ -111,9 +111,9 @@ evals-lab run <bundle-dir | pipeline.yaml> [--items <dir>] [--json <file|->]
 | `--min-pass` | an eval passes when this share of its items pass |
 | `--progress` | a line per item as it finishes |
 
-The bundle holds no key. Each Target profile's key comes from
+The bundle holds no key. Each AI Connection's key comes from
 `EVALSLAB_API_KEY_` and its slug in Setup, in capitals with hyphens as
-underscores: a profile slugged `anthropic` reads `EVALSLAB_API_KEY_ANTHROPIC`,
+underscores: a connection slugged `anthropic` reads `EVALSLAB_API_KEY_ANTHROPIC`,
 and `gpt-4o-mini` reads `EVALSLAB_API_KEY_GPT_4O_MINI`. A table of results
 goes to stderr.
 
@@ -237,6 +237,12 @@ npm install -g evals-lab@latest   # upgrade
 npm uninstall -g evals-lab        # remove; your data directory stays
 ```
 
+## Contact
+
+[evals-lab@grainbox.com](mailto:evals-lab@grainbox.com) for questions,
+problems and feedback. More at
+<https://www.grainbox.com/products/evals-lab/>.
+
 ## Licence
 
 Copyright 2026 Grainbox Limited, under the

package/bin/run.js

@@ -62,7 +62,7 @@ const USAGE = `Usage: evals-lab run <bundle-dir | pipeline.yaml> [options]
                      --progress need it.
   --help             this.
 
-A bundle's Target profile keys come from $EVALSLAB_API_KEY_<SLUG>. Exit
+A bundle's connection keys come from $EVALSLAB_API_KEY_<SLUG>. Exit
 codes: 0 pass (or queued, without --wait), 1 an eval failed, 2 incomplete (an
 item missing, one that did not run, or the run cancelled), 3 broken or
 refused.`;
@@ -392,16 +392,20 @@ async function runOnLab(core, o, { env, stdout, say }) {
   }
   const call = labApi(base, env, workspace);
 
-  // What the page reads before it submits: the lab's pipelines and Target
-  // profiles, the Source the pipeline reads and the eval groups it may link.
+  // What the page reads before it submits: the lab's pipelines, AI
+  // Connections and Connection Profiles, the Source the pipeline reads and
+  // the eval groups it may link. Slugged, as RunsTab's submit has them, so
+  // the run carries the slug its keys are spelt from (#253).
   const docs = (await call("GET", "/api/state")).docs ?? {};
-  const setup = docs["promptlab.profiles"]?.body ?? {};
-  const profiles = Array.isArray(setup.list) ? setup.list : [];
+  const setup = docs["promptlab.connections"]?.body ?? {};
+  const connections = core.withSlugs(Array.isArray(setup.list) ? setup.list : []);
+  const held = docs["promptlab.connectionProfiles"]?.body?.list;
+  const answers = core.answersOf(connections, Array.isArray(held) ? held : []);
   const saved = docs["promptlab.workflows"]?.body?.list;
   const found = pipelineNamed(Array.isArray(saved) ? saved : [], o.pipeline, base);
-  const types = new Map(profiles.map(p => [p.id, p.type]));
-  const doc = core.upgradePipeline(found.work, { profileType: id => types.get(id) });
-  const g = profiles.find(p => p.id === setup.grader);
+  const doc = core.upgradePipeline(found.work, { profileType: id => answers.connection(id)?.type,
+                                                 connectionSplit: answers.split });
+  const g = answers.connection(setup.grader);
   const grader = g ? { id: g.id, name: g.name } : null;
   const content = core.contentOf(doc);
   let source = null;
@@ -411,18 +415,14 @@ async function runOnLab(core, o, { env, stdout, say }) {
   const files = source ? source.files.map(f => f.name) : undefined;
   const { datasets } = await call("GET", "/api/datasets");
   const problems = core.validatePipeline(doc, {
-    profiles: id => profiles.find(p => p.id === id) ?? null,
+    profiles: answers.resolve, connections: answers.connection,
     sources: id => (source && id === content.ref.id ? { name: source.name, type: source.type, files } : null),
     grader, groups: datasets ?? [],
   });
   if (problems.length) throw new Refused(`${found.name} cannot run: ${problems.join("; ")}`);
-  // As RunsTab's submit resolves it: slugged, so the run carries the slug
-  // its keys are spelt from (#253), and keyless -- the lab fills each key in
+  // As RunsTab's submit resolves it: keyless -- the lab fills each key in
   // from its own store.
-  const slugged = core.withSlugs(profiles);
-  const run = core.resolvePipeline(doc, {
-    profiles: id => slugged.find(p => p.id === id) ?? null, grader, ...(files ? { files } : {}),
-  });
+  const run = core.resolvePipeline(doc, { profiles: answers.resolve, grader, ...(files ? { files } : {}) });
 
   const queued = (await call("POST", "/api/queue", run)).run;
   say(`evals-lab run: ${found.name} is queued on ${base} as run ${queued.id}`);

package/lab/VERSION

@@ -1 +1 @@
-0.10.0 (2026.10.09-614)
+0.11.0 (2026.10.11-671)

package/lab/demo/pipelines/demo-1.json

@@ -1,5 +1,5 @@
 {
-  "version": 15,
+  "version": 16,
   "id": "demo-1",
   "name": "Demo 1",
   "jobs": [
@@ -105,10 +105,11 @@
     {
       "id": "s1",
       "name": "",
-      "profile": {
+      "connection": {
         "id": "vision",
         "name": "Vision"
       },
+      "profile": null,
       "steps": [
         {
           "type": "prompt",
@@ -119,10 +120,11 @@
     {
       "id": "s2",
       "name": "",
-      "profile": {
+      "connection": {
         "id": "vision",
         "name": "Vision"
       },
+      "profile": null,
       "steps": [
         {
           "type": "prompt",

package/lab/demo/pipelines/demo-2.json

@@ -1,5 +1,5 @@
 {
-  "version": 15,
+  "version": 16,
   "id": "demo-2",
   "name": "Demo 2",
   "jobs": [
@@ -105,10 +105,11 @@
     {
       "id": "s1",
       "name": "",
-      "profile": {
+      "connection": {
         "id": "vision",
         "name": "Vision"
       },
+      "profile": null,
       "steps": [
         {
           "type": "prompt",
@@ -119,10 +120,11 @@
     {
       "id": "s2",
       "name": "",
-      "profile": {
+      "connection": {
         "id": "vision",
         "name": "Vision"
       },
+      "profile": null,
       "steps": [
         {
           "type": "prompt",