Sploink logo for the AI agent observability platformSploink
Private beta launching Q1 2026

Low-level agent telemetry at the execution layer. Know when your agent is making progress or stuck in a loop. Datadog for agents.

Sploink — refactor the auth module
claude-code47 events4m 12s
Live

Convergence

0.74

Thrash

STABLE

Momentum

+0.12 ↑

Mode

OnTrack

Recent Events

file_editedauth.py
+23 lines
command_finishedpytest
exit 0
test_finished14/14 passed
100%
git_diff_checkeddiff 1.2kb
growing
file_editedmiddleware.py
+8 lines
commit_createdmain
a3f8c21

Convergence Timeline

thrash
0events47

Breakdown

52%
Git
31%
Test
17%
Churn

Live Metrics

Progress0.68
Drift0.14
Similarity0.22
Cost0.31
Scroll to explore

50+

metrics per execution

100ms

event capture latency

40%

of agent time is thrash

2024

year eBPF monitoring launches

Agent Observability

Know what your agents are really doing

Every observability tool traces API calls. Sploink traces execution behavior. See convergence, detect thrash, measure momentum.

Convergence Score

Real-time measurement of whether your agent is moving toward the goal or standing still.

Thrash Detection

Automatically detect when agents are stuck in loops, editing the same files repeatedly without progress.

Execution Layer Visibility

See what agents actually do at the system level—not just API calls, but every command and file change.

Git State Tracking

Monitor diff growth, commit history, and real work accumulation over time.

Multi-Agent Comparison

Compare agent performance across tasks and identify which agents excel at specific job types.

Momentum Curves

Visualize velocity of real work over time. Healthy sessions ramp. Stuck sessions flatline.

Test & Build Tracking

Track test runner output and build logs to see if the codebase is getting closer to working.

Framework Agnostic

Works with Claude Code, Cursor, GPT-4o, and any open-source agent framework.

Convergence Score

Real-time measurement of whether your agent is moving toward the goal or standing still.

Thrash Detection

Automatically detect when agents are stuck in loops, editing the same files repeatedly without progress.

Execution Layer Visibility

See what agents actually do at the system level—not just API calls, but every command and file change.

Git State Tracking

Monitor diff growth, commit history, and real work accumulation over time.

Multi-Agent Comparison

Compare agent performance across tasks and identify which agents excel at specific job types.

Momentum Curves

Visualize velocity of real work over time. Healthy sessions ramp. Stuck sessions flatline.

Test & Build Tracking

Track test runner output and build logs to see if the codebase is getting closer to working.

Framework Agnostic

Works with Claude Code, Cursor, GPT-4o, and any open-source agent framework.

Three Core Metrics

Know what agents actually do

Three simple metrics that tell you everything about agent execution. Real-time visibility at the behavior level.

Convergence

Is the agent actually getting closer to done, or just spinning its wheels?

Thrash

Catches when the agent is stuck in a loop — editing the same files, running the same commands, going nowhere.

Momentum

Shows how fast real work is happening over time — rising means progress, flat means stuck.

CONVERGENCE
40%
Git state weight
THRASH
Real-time
Loop detection
MOMENTUM
Velocity
Over time tracking

Real-time Feedback

See agent behavior as it happens

Run sploink init once to install the Claude Code plugin. Every session after that is automatically tracked — convergence, thrash patterns, and momentum, without changing how you work.

  • Native Claude Code plugin — no wrapper needed
  • Live convergence scoring dashboard
  • Real-time thrash detection
  • One-time setup with sploink init
Terminal

Zero-config telemetry

Run agents with observability

Wrap any agent command and get real-time convergence scoring, thrash detection, and momentum curves. Works with Claude Code, Cursor, GPT-4o, and any framework.

terminal
Documentation

Understand agent behavior

Learn how to run agents with Sploink, interpret telemetry, and optimize for convergence.

Quick Start

Install Sploink and run your first agent session in 5 minutes

Agent Frameworks

Guides for Claude Code, Cursor, GPT-4o, open-source agents

CLI Reference

Complete sploink command reference and configuration

Metrics & Scoring

Understanding convergence scores, thrash detection, momentum curves

HTTP API

POST /execute endpoint and telemetry ingestion schema

Architecture

Deep dive into Sploink's design and long-term roadmap

View full documentation→
Enterprise

Agent infrastructure at scale

Sploink powers agent platforms, marketplaces, and orchestration systems. Build the agentic OS layer.

eBPF Kernel Probes

Kernel-level visibility into memory, syscalls, I/O patterns, and process trees.

Agent Reputation Graph

Multi-agent comparison with historical performance data across thousands of sessions.

Dedicated Support

Direct access to the Sploink team for integration and custom metric design.

Marketplace Ready

Build defensible agent businesses using proven behavioral telemetry data.

On-Premise Deployment

Deploy Sploink in your own infrastructure for complete data control.

Custom Metrics

Design proprietary metrics and scoring systems tailored to your agent types.

Salaxia Integration

Deep integration with agent orchestration and routing platforms.

Custom Dashboards

White-labeled dashboards and reporting for your agent marketplace.

Building the future of agents?

Talk to our team about Sploink for your platform, marketplace, or infrastructure.

Stop guessing about agent progress.

Know exactly what your agents are doing — in real time. See convergence, detect loops, measure momentum.