Get more out of your AI.

AgentOps reduces repeated work, optimizes what reaches your agents, and helps you improve the way you work with them across every session.

Try it $ npm i -g https://useagentops.com/download/agentops-latest.tgz
Ready in about 47 seconds
Open source · Local by default · Numbers only
Your code never leaves your machine.
Without AgentOpsraw
❯ npm test
Compiling… ok · running 96 tests
test policy::resolves_nested … ok
test cache::dedupes_stdout … ok
…214 more lines of passing output
Read how-to-code.md (again)
Read how-to-code.md (again ×3)
curl api… #1 #2 #3 #4 #5 #6 #7 #8 #9
{ "id":"0x…", "revision":"7", "meta":{…}}
…full JSON body, every turn
context92% full · re-sent each turn
With AgentOpsshaped
❯ npm test
✓ 96 passed, 0 failed · 2.31s
 
how-to-code.md → cached (read once)
 
9 identical calls → flagged: batch these
 
JSON → id, revision, status
 
signal kept in context, noise dropped
contextlean · signal only

Same session. AgentOps shapes supported tool results before they fill the context window, reduces repeated work, and shows you where unnecessary repetition occurred.

~12,861 → ~461 tokens on one command (estimated) 0 duplicate reads Less context re-sent per turn
What it does

Four ways it earns its place

No dashboard to monitor. AgentOps works alongside your agents and surfaces what matters when you need it.

Reduce repeated work

Agents often re-read the same files and repeat the same calls during a session. AgentOps sees that repetition and shows you where you're repeating work, so you can cut it.

how-to-code.md read → flagged

Optimize what reaches your agents

Supported tool results are shaped locally before they reach the context window. Your agent reasons over the useful signal instead of carrying unnecessary raw output through later turns.

raw ~12,861 → shaped ~461 tokens (estimated)

See how you operate

Patterns across your sessions and agents become clear feedback. See where time and tokens go, what repeats, and which operating patterns are worth reviewing.

9 serial calls → consider batching

Works across your agents

Use one installation across Claude Code, Codex, Gemini and OpenCode. Get a consistent view of how your agents behave wherever the work happens.

1 install · multiple agents
Privacy by architecture

Your code never leaves your machine

Shaping and analysis happen locally. Standard submissions contain numbers only, and they are sent only when you choose to submit. Any richer diagnostic sharing will always require separate, explicit consent.

Your machine

Raw output and code

Tool results are captured and shaped locally. Source code and raw output are not included in a standard submission.

numbers
only →
Optional submission

Scores and counts

Token counts, grades and aggregate measurements. No code, file paths or raw output.

Open source. Read the code ↗ Local by default. Nothing leaves until you say so.
Works with

The agents you already run

Claude Code
Anthropic
Alpha
Codex CLI
OpenAI
Alpha
Gemini CLI
Google
Alpha
OpenCode
 
Alpha
Cursor
Coming soon
Windsurf
Coming soon
Aider
Coming soon

Free and open. Install, connect, and use it without limits.

Get more out of your AI.

One command. AgentOps detects your supported agent and starts working in the current session.

Try it $ npm i -g https://useagentops.com/download/agentops-latest.tgz
Ready in about 47 seconds
Open source · Local by default · Numbers only. Your source code never leaves your machine.