Infrastructure · Early access

Your AI agents are doing more than generating responses.

Rethen connects agent activity across models, tools, workflows, and external systems so your team can follow the work and understand what happened next.

Infrastructure for the next generation of autonomous software.

Investigation — Agent Run #8421
EXEC

Agent Run #8421

10:01:24 · 18 LLM calls · 7 tool calls

LLM

18 LLM calls

GPT-4o, Claude 3.5 · 82k tokens

TOOL

7 tool calls

GitHub, Jira, Search

SYS

4 systems touched

GitHub · Jira · CI/CD · Deploy API

GH

GitHub PR #482

Merged 10:13:08 · 3 executions

DEPLOY

Deployment #193

api-gateway v2.1.4 · 10:18:47

INC

Production Incident #91

42 connected events · 3 executions

hover or click nodes to inspect

Market context

Agents are moving from answering questions to doing work.

40%

up from 27%

of organizations with >$1B revenue scaling AI agents

McKinsey State of AI, 2026

74%

expecting to use agentic AI at least moderately within two years

Deloitte State of AI, 2026

21%

have mature governance

of surveyed organizations report mature governance for autonomous AI agents

Deloitte State of AI, 2026

150k+

projected avg. agents at a Fortune 500 enterprise by 2028

Gartner, 2024

More agents deployed means more autonomous work, more systems touched, and more outcomes to account for. The question of what happened is getting harder to answer.

The problem

The trace ends before the story does.

What execution tracing shows

ST

Agent

MO

LLM

TO

Tool

TO

Tool

EN

Response

The trace captures what the agent did during execution.

What actually happened

ST

Agent

MO

LLM calls

TO

Tools

GH

GitHub

AR

PR #482

CI

CI pipeline

DE

Deployment #193

SY

Production

IN

Incident #91

Production systems need you to understand what came after.

Questions teams are left with

Q01

Which AI executions were connected to this failed deployment?

Q02

What did this agent change before this incident appeared?

Q03

Which tools and systems did this workflow touch?

Q04

What happened after this agent completed?

Q05

Can we reconstruct the chain later?

Q06

Who or what triggered this sequence of events?

How Rethen fits

Observability shows the execution.
Rethen follows the work.

AI Observability

Agent

LLM

Tool

Response

Traces and logs of what happened inside the agent execution. Spans, tokens, latency, tool calls.

Rethen

Agent

Execution

Tools

External systems

Entities

Outcomes

Connects execution data with the systems, entities, and outcomes around it.

Rethen is not trying to replace your existing tracing stack. It connects execution data with the systems and outcomes around it — so you can move from “the trace ended” to “this is what happened next.”

How it works

Connect the pieces your existing systems keep separate.

01

Collect

Bring together activity from LLMs, agents, tools, OpenTelemetry, GitHub, Jira, deployments, and webhooks.

LLMsAgentsToolsOpenTelemetryGitHubJiraDeploymentsWebhooks
02

Resolve

Rethen builds relationships between events and entities across systems that use different identifiers.

Agent Run #8421

PR #482

Deployment #193

03

Connect

One piece of AI work can connect to many events, and one outcome can connect back to multiple executions.

3 agent executions

1 PR

1 deployment

1 incident

04

Investigate

Start from wherever the question begins and follow the chain — forward from execution or backward from outcome.

Execution

PR

Deployment

Incident

Product interface

Start anywhere. Follow the chain.

An incident surfaces at 11:42 AM. Rethen traces it back through the deployment, the PR, and the agent run that created it.

Investigation/Incident #91
active
INC

Incident #91 · Production deployment failure

11:42 AM · Detected automatically · P1

Connected activity

EXEC

Agent Run #8421

Started 10:01:24

GH

GitHub PR #482

Merged 10:13:08

DEPLOY

Deployment #193

api-gateway v2.1.4

INC

Production Incident #91

11:42 AM · active

Detail view

click a node to inspect

Bidirectional

Investigate from the AI or from the outcome.

Production questions rarely start in the same place. Rethen lets teams move through the chain in either direction.

Starting from AI
EXECAgent
EXECExecution
TOOLTool calls
GHGitHub
DEPLOYDeployment
OUTOutcome
Starting from outcome
INCIncident
DEPLOYDeployment
GHPR
EXECAgent execution
TOOLTool calls
LLMLLM activity

Infrastructure

The graph underneath the activity.

Rethen treats agent activity as a connected system of events and entities — not isolated logs.

Entity relationship model

executedtriggeredattributedcreatedAgentExecutionLLM callToolPRDeploymentIncidentTicket

Entities

Agent
Execution
LLM call
Tool
Repository
PR
Ticket
Deployment
Incident

Relationships

executed
called
created
updated
triggered
connected
attributed

Every event in Rethen is typed and linked. An agent run doesn't just exist — it executed, called tools, and created artifacts that triggered downstream events.

The graph model is what makes historical replay and bidirectional investigation possible — relationships are explicit, not inferred at query time.

Historical replay

Your investigation should not depend on when you noticed the problem.

Rethen keeps raw activity separate from derived relationships, allowing historical analysis to be recalculated as your understanding changes.

How replay works

01

Raw events

Immutable activity log — never rewritten

02

Relationship rules

Versioned, updatable attribution logic

03

Attribution

Derived graph — recalculated on rule change

Rule version comparison

Rule v1repo match only
82 outcomes
Rule v2repo + time window
117 outcomes

+35 additional outcomes attributed by refining the rule

Raw event storage is separate from the derived relationship graph. When attribution rules improve, the graph is recalculated against the full historical dataset — no data is lost and no manual backfilling is required.

What you get

Know what your agents actually did.

01

Follow AI work across systems.

Trace agent activity past the execution boundary and into GitHub, CI/CD, deployments, and other external systems.

02

Investigate incidents faster.

Start from an alert or incident and navigate directly to the agent executions and tool calls that preceded it.

03

Connect AI activity with real outcomes.

Link autonomous work to engineering and business entities — PRs, deployments, tickets, incidents — not just trace spans.

04

Keep the history you'll need later.

Raw events are stored separately and durably. Investigations can happen hours, days, or weeks after the fact.

05

Recalculate when rules change.

Attribution rules are versioned. When your understanding of the system improves, replay the full history against updated logic.

06

Build on your existing stack.

Rethen connects to what you already have — OpenTelemetry, GitHub, Jira, deployment pipelines — without replacing them.

Use cases

Built for teams deploying agents into real systems.

Engineering

Understand what AI changed before production changed.

When a deployment breaks production, reconstruct every agent execution that contributed — from model calls to the PR that triggered the deploy.

Activity chain

EX

Agent

GH

PR #482

CI

CI pipeline

DE

Deployment #193

IN

Incident #91

Integrations

Built to sit across your existing stack.

Rethen connects to the tools you already use. Below is an honest view of what works today and what is on the roadmap.

Available

OpenAI

LLM

Anthropic

LLM

OpenTelemetry

Tracing

GitHub

VCS

Jira

Tracking

Webhooks

Generic

Planned

Google Gemini

LLM

GitLab

VCS

Linear

Tracking

PagerDuty

Incidents

Datadog

Monitoring

AWS CDK

Deploy

Audience

For teams putting agents into real systems.

AI

AI & ML Engineers

Understand agent executions beyond model traces. See what systems your agents touched and what changed.

PLAT

Platform Engineers

Create a common observability layer across agents, infrastructure, and the external systems they touch.

ENG

Engineering Leaders

Investigate AI-related production events without manually stitching together logs from five different systems.

ORG

AI-First Companies

Build a historical record of autonomous software activity that scales as your deployments grow.

Early stage

We are early.

Rethen is being built around a problem we believe will become more important as autonomous software moves deeper into production.

We are currently building the infrastructure, testing the model with real workflows, and learning where teams feel the problem most strongly.

Talk to us about how you're tracking agent activity

Where we are today

Core event collection infrastructure

In progress

Relationship resolution engine

In progress

GitHub & OpenTelemetry integration

In progress

Investigation interface

Prototype

Historical replay & backfill

Planned

Attribution rule versioning

Planned

Multi-tenant deployment

Planned

Early access

Know what your agents did.

Follow the work. Understand the outcome.

Early access to the investigation interface

Ability to connect your agent workflows

Direct conversations with the team building Rethen

No spam. We read every submission.