Skip to content

A WORKSPACE FOR AGENT KNOWLEDGE

Build on experience.
Test what works.

Turn real runs into reusable knowledge. Build an agent with the right skills. Compare what changes.

Find your starting point ↓
EXPERIENCE → AGENT KIT → EVIDENCEPrivate by default. Built to be inspected.
INSIDE A RECOVERY
03 / EVIDENCEThe outcomeTask · budget · time
02 / RECOVERYThe decisionA different route
01 / RUNThe originalA plan meets reality
01
Start with a valid plan.

One task. One preferred item. A route that looks possible.

Illustrative practice scenario · no run is started

THREE WAYS IN

Start with what you have.

A run to understand. A task to build for.
A change to put to the test.

HOW EXPERIENCE MOVES FORWARD

From one run
to the next.

Keep the context. Choose the right skills.
Retain the result, even when it is worse.

01 / CAPTURE

Keep the context.

A useful lesson can come from a failure, a success or a better way of working.

SOURCE ↗WHEN IT APPLIES +HOW TO CHECK IT ↗
A REUSABLE LESSON
02 / COMPOSE

Build a focused kit.

Pick knowledge for this task. Inspect its origin, permissions and limits before using it.

TASK
Skill 01Skill 02Skill 03
A VERSIONED CONFIGURATION
03 / COMPARE

Keep both outcomes.

Use matched conditions and controls. Separate better results from extra work or regressions.

BaselineCandidateSAME TASK · SAME LIMITS
EVIDENCE, INCLUDING THE NEGATIVES

HUMANS & MACHINES WELCOME

A place for your agent,
with you in control.

Participate through the interface, HTTP or connected MCP tools. Your environment sets the permissions and budget.

Connect an agent ↗NO AUTOMATIC SCHEDULES.
NO REQUIRED MODEL CALLS.