Skip to content

Commit 7c8172e

Browse files
committed
docs: add Evals page under Agent Settings; link it in the overview recap
1 parent 379e307 commit 7c8172e

2 files changed

Lines changed: 64 additions & 1 deletion

File tree

os/agent-settings/evals.md

Lines changed: 60 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,60 @@
1+
# Agent Settings: Evals
2+
3+
![Evals panel in the Edit Agent modal showing the benchmark selector, Manage benchmarks and New Evaluation buttons, and a table of evaluation runs with Status, Initiated By, Created, and Actions columns](/images/docs/os/agent-settings/agent_settings_evals.webp)
4+
5+
## Overview
6+
7+
The Evals panel measures how well an agent performs by running it against a **benchmark** — a set of test questions — and scoring every response, so you can catch weak spots and track improvements as you refine the agent. The panel header reads: "Run this agent against a benchmark and score the responses," and the helper banner explains: "Measure how well your agent performs. Run it against a benchmark — a set of test questions — and every response gets scored, so you can catch weak spots and track improvements as you refine the agent."
8+
9+
A benchmark is a graded test set: each item pairs an **input** (the question) with an optional **expected output** (the answer you want). These evaluation benchmarks are separate from an agent's [Datasets](datasets.md) — Datasets are the RAG documents that ground the agent's answers, while benchmarks are the test cases used to grade them. Benchmark items can be authored directly, uploaded from a CSV, or seeded from existing chat traces so real conversations become test cases.
10+
11+
Running an evaluation is asynchronous: it dispatches a background job that sends every benchmark question to the agent and records each response as a **trace**. A run moves through `pending → in progress → completed`, and must reach **Completed** before it can be reviewed or exported.
12+
13+
To reach this screen, open the **Edit Agent** modal, go to the **Analytics** tab, and select **Evals** from its sidebar (alongside Memory, History, and Audit). Evals is admin-gated: it is only available to users with permission to view the agent's analytics. Evaluation data is scoped to your organization and isolated — no other organization can read your benchmarks, runs, or scores.
14+
15+
## Target Audience
16+
17+
**Administrator** | **Agent Builder**
18+
19+
## Running an Evaluation
20+
21+
The panel lists every evaluation run for the agent in a table with **Evaluation**, **Status**, **Initiated By**, **Created**, and **Actions** columns. Above the table, a benchmark selector chooses which test set to run, **Manage benchmarks** opens benchmark authoring, and **New Evaluation** starts a run of the selected benchmark against the current agent.
22+
23+
Each row's **Actions** menu (``) exposes:
24+
25+
#### View results
26+
Opens the run detail — every trace with its input, expected output, actual output, and any scores.
27+
28+
#### New review
29+
Starts an LLM-as-Judge review of the run (see below).
30+
31+
#### Check status
32+
Polls the background task's current state (`pending`, `in progress`, `completed`, or `failed`).
33+
34+
#### Export CSV
35+
Downloads the run's results as a spreadsheet — one row per item with `item_id`, `input`, `expected_output`, `actual_output`, `trace_id`, and one column per score.
36+
37+
#### Delete
38+
Removes the run. This is destructive and is confirmed before it proceeds.
39+
40+
## Reviews (LLM-as-Judge)
41+
42+
![Evaluation run detail showing raza-eval-2 marked Completed, 3 traces produced, a Reviews section with a completed accuracy review scored by gpt-5, and the benchmark items listed with their score counts](/images/docs/os/agent-settings/agent_settings_evals_experiment.webp)
43+
44+
Opening a completed run shows its header — benchmark name, when it started, and how many traces it produced — above a **Reviews** section. As the panel explains: "Reviews are LLM-as-Judge runs that score every trace in this experiment. Scores from completed reviews appear under each item below."
45+
46+
A review grades every response in the run automatically. Selecting **New review** defines a **criteria** rubric in plain language (for example, "Evaluate the response on accuracy") and picks the judge model; a separate LLM then reads each item's input, expected output, and the agent's actual output and assigns a score. Each review shows its status, the judge model that ran it (for example, `gpt-5`), and when it finished. Completed review scores roll up under each benchmark item, so an item can carry several scores from different reviews.
47+
48+
![Expanded accuracy review showing the criteria field and four produced scores, each with a numeric value and the judge's written reasoning for that benchmark item](/images/docs/os/agent-settings/agent_settings_evals_review.webp)
49+
50+
Expanding a review reveals its **Criteria** and the **scores produced** — one per benchmark item, each with a numeric value and the judge's written reasoning. In the example above, the *accuracy* review explains why a clear, correct answer earned a full score while an empty or unanswerable response received a lower, neutral one. Reviews are the automated grading path; individual human, numeric, boolean, or categorical annotations can also be attached to any trace, and reusable rubrics can be saved and referenced across runs.
51+
52+
## Programmatic Access
53+
54+
Everything in this panel is available over the platform API, so evaluations can run in CI or from an AI agent: create benchmarks and items (JSON, CSV upload, or from chat traces), start experiment runs, launch LLM-as-Judge reviews, read scores, and export CSV. The API is packaged as the `iblai-api-agent-eval` skill at [github.com/iblai/api](https://github.com/iblai/api).
55+
56+
## Related
57+
58+
- [Agent Settings: Analytics](analytics.md) — usage reports and dashboards for the agent
59+
- [Agent Settings: Chat History](history.md) — the raw conversation traces evaluations can be seeded from
60+
- [Agent Settings: Datasets](datasets.md) — the RAG documents that ground the agent (distinct from evaluation benchmarks)

os/overview.md

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -33,6 +33,9 @@ Extend agent capabilities with Model Context Protocol tool servers, configured p
3333
#### Analytics
3434
Usage dashboards, per-user activity, topic analysis, a transcript viewer, cost tracking, and exportable data reports. See the [Analytics section](analytics/overview.md).
3535

36+
#### Evaluations
37+
Measure and improve agent quality: run an agent against a benchmark of test questions, then score every response with LLM-as-Judge reviews and human annotations, and export the results. See [Evals](agent-settings/evals.md).
38+
3639
#### Multi-Organization, SSO & RBAC
3740
Full organization isolation with per-organization configuration and branding, Single Sign-On, and granular role-based access control with roles, policies, groups, and teams. See [Organization Settings](organization-settings/organization.md).
3841

@@ -48,7 +51,7 @@ Embed agents in any website via iframe with custom styling, and integrate progra
4851
The end-user workspace: [Chat Interface](chat-canvas/chat.md) · [Canvas](chat-canvas/canvas.md) · [Canvas Options](chat-canvas/canvas-options.md) · [Downloading a Canvas](chat-canvas/canvas-download.md) · [Renaming a Canvas](chat-canvas/canvas-rename.md)
4952

5053
#### Agent Settings
51-
Everything configurable on a single agent: [Basic](agent-settings/basic.md) · [System Prompt](agent-settings/prompt.md) · [LLM Selection](agent-settings/llm-selection.md) · [LLM Configuration](agent-settings/llm-configuration.md) · [Capabilities](agent-settings/capabilities.md) · [Tools](agent-settings/tools.md) · [Skills](agent-settings/skills.md) · [MCP Servers](agent-settings/mcp.md) · [Datasets](agent-settings/datasets.md) · [Memory](agent-settings/memory.md) · [Voice](agent-settings/voice.md) · [Voice Selector](agent-settings/voice-selector.md) · [Voice Calls](agent-settings/voice-call.md) · [Safety](agent-settings/safety.md) · [Privacy](agent-settings/privacy.md) · [Disclaimers](agent-settings/disclaimers.md) · [Discovery](agent-settings/discovery.md) · [Access Control](agent-settings/access.md) · [Embed](agent-settings/embed.md) · [API Access](agent-settings/api.md) · [Agent Analytics](agent-settings/analytics.md) · [Audit Log](agent-settings/audit.md) · [Chat History](agent-settings/history.md) · [Tasks](agent-settings/tasks.md) · [Sandbox](agent-settings/sandbox.md) · [LTI Keys](agent-settings/lti-keys.md) · [LTI Links](agent-settings/lti-links.md) · [LTI Tool Endpoints](agent-settings/lti-tool-endpoints.md) · [LTI Tools](agent-settings/lti-tools.md)
54+
Everything configurable on a single agent: [Basic](agent-settings/basic.md) · [System Prompt](agent-settings/prompt.md) · [LLM Selection](agent-settings/llm-selection.md) · [LLM Configuration](agent-settings/llm-configuration.md) · [Capabilities](agent-settings/capabilities.md) · [Tools](agent-settings/tools.md) · [Skills](agent-settings/skills.md) · [MCP Servers](agent-settings/mcp.md) · [Datasets](agent-settings/datasets.md) · [Memory](agent-settings/memory.md) · [Voice](agent-settings/voice.md) · [Voice Selector](agent-settings/voice-selector.md) · [Voice Calls](agent-settings/voice-call.md) · [Safety](agent-settings/safety.md) · [Privacy](agent-settings/privacy.md) · [Disclaimers](agent-settings/disclaimers.md) · [Discovery](agent-settings/discovery.md) · [Access Control](agent-settings/access.md) · [Embed](agent-settings/embed.md) · [API Access](agent-settings/api.md) · [Agent Analytics](agent-settings/analytics.md) · [Evals](agent-settings/evals.md) · [Audit Log](agent-settings/audit.md) · [Chat History](agent-settings/history.md) · [Tasks](agent-settings/tasks.md) · [Sandbox](agent-settings/sandbox.md) · [LTI Keys](agent-settings/lti-keys.md) · [LTI Links](agent-settings/lti-links.md) · [LTI Tool Endpoints](agent-settings/lti-tool-endpoints.md) · [LTI Tools](agent-settings/lti-tools.md)
5255

5356
#### Analytics
5457
Organization-wide insight: [Overview Dashboard](analytics/overview.md) · [Users](analytics/users.md) · [Topics](analytics/topics.md) · [Transcripts](analytics/transcripts.md) · [Costs](analytics/costs.md) · [Data Reports](analytics/data-reports.md) · [Audit](analytics/audit.md)

0 commit comments

Comments
 (0)