Runtime evidence for AI-assisted development

Code changes outrun our understandingSee how every change behaves before it merges

AppMap records and compares application behavior. Recording happens locally, in your environment, and the key traces are committed to the repository with the code. Coding agents query those traces and verify their changes against them. Developers inspect the same behavior as diagrams.

main · gold_traces/
the baseline
appmap-review
monitoring with Gold Traces
agents/frontend
coding
agents/paid-api
coding
38 Gold Traces on main
blocked before merge
1 changed, 1 new
baseline advances
behavior held
v1.4.0, baseline frozen
Coding agents work the branches. The monitor compares every change against the Gold Traces on main. The release freezes the set.

With AI coding agents, code is shipping fast. Confidence is not.

AI writes more of the code every month. Developers and coding agents still have to work out what each change will do when it runs.

  • 96%

    of developers do not fully trust AI-generated code

    Sonar, 2026
  • 70%

    of software leaders say quality has already degraded as AI sped up delivery

    SmartBear, 2026
  • 60%

    hit quality issues last year because code outran testing

    SmartBear, 2026

Catch unexpected behavior while the change is still in the editor.

AppMap gives developers and coding agents runtime evidence as they work, so they can investigate calls, queries, exceptions, and side effects before the change merges.

Accelerate review and acceptance of AI code changes

AppMap runs locally in development and centrally in the pull request flow and CI, across large codebases and many teams. Reviewers, coding agents, and compliance teams see the same record of what each change did, so organizations can accept changes faster.

execution pathsauth + securityAPI contractdata flowbusiness logicinteractionspersistence + SQLconcurrencyside effectsfailure handlingperformanceAI-06 · catch-all swallows real failuresAI-11 · retry now doubles the side effectAI-09 · eager load flipped to lazy
Find the defects in the interactions.

See the full runtime behavior analysis, with the standards it maps to →

What developers say about AppMap

Rated five stars by developers across the Visual Studio Code and JetBrains marketplaces, from a community of more than 100,000 developers.

★★★★★

"Helped me better understand and analyze the hierarchy and structure of my applications. Looks awesome, performs great, and provides truly invaluable feedback."

Max, Java developer, VS Code Marketplace
★★★★★

"An awesome plugin for sequence diagrams."

Shaun Wang, Java developer, JetBrains Marketplace
★★★★★

"AppMap improves productivity on both fronts, coding and documentation. It has become an inseparable part of our development process."

Venkat Bagam, Cognitive Zen

Review behavior before it ships.

Keep the code review. Add a review of what the change did when it ran. AppMap compares fresh traces with the base revision and shows what held and what changed. Developers inspect the diagrams. Coding agents query the same traces. The findings can go straight to the pull request.

APPMAP BEHAVIORAL REVIEW
head working treebase 8e54ce2 (v0.21.0 baseline)traces 41 · new 2 · changed 0
MEDIUMFIXED DURING REVIEW

A real review, from a production application maintained mostly by AI agents

Two code paths wrote the same record. Each was correct alone. Every test passed.

Request handlerBackground jobread record → owner = Ajob updates the recordCOMMIT owner = Ccheck owner == A ✓ (stale)COMMIT owner = Bthe job's write is silently erased
FIX, APPLIED AND RE-RECORDED

The handler now locks the row before the ownership check. The recorded SQL confirms it: the read is now SELECT … FOR UPDATE.

RE-RECORDING CHANGED ONE TRACE

1 trace changed by the fix. The other 39 stayed byte-identical.

behavior the change did not touch: none moved ownership checked before and after the write, visible in the traces 4 new queries, all parameterized
gold traces 41 · server tests 427 passed
AppMap dependency map of a running application, generated from recorded traces.
Developers open the same traces as interactive diagrams in the editor.

See what every change does.

AppMap runs in development and CI, before code merges. Each change produces fresh traces. AppMap turns them into maps developers can inspect to see what ran, what the change touched, and where behavior moved.

AppMap call tree showing the request path through the application.
The call treeThe request path the agent queried.
AppMap view of the SQL queries executed during a request, shown as parameterized statements.
The queriesThe exact SQL it ran, query by query.
AppMap code map of functions, APIs, and objects captured in the trace.
The code objectsFunctions, classes, and queries captured in the trace.

Three representative views. Explore the broader visual set in your editor before the change merges.

Does runtime context improve AI coding performance?

In one internal study, yes. Under a three-tool-call limit, agents using AppMap reached 100% root-cause accuracy. Source-only agents reached 28%. For an organization running agents across many repositories, that means fewer investigations ending without a diagnosis.

In the primary study, the source-only condition cost about 3.4 times as much.

The lowest-cost configuration that solved every primary-study case was a compact model reading traces.

One internal study, Claude models. Method and limits in the benchmark.

Developers trust AppMap

~140K
editor installs
100K+
developer community
2020
open source since

Go deep before you merge.

Production and canary tracing are built to observe systems continuously with limited overhead. AppMap records deeply in development, capturing the code-level execution path, queries, HTTP calls, exceptions, and relationships behind the recorded runs. The result is information-dense context developers can inspect and coding agents can query before the change merges.

Canary / post-deploy

Broad, lightweight telemetry

  • API
  • Service A
  • Service B
  • Database
sampled spansservice boundariesproduction-safe detail

AppMap in development

Deep execution context

POST /charge  PaymentController#charge    ChargeService#authorize      RetryPolicy#wrap        LedgerService#write          SQL INSERT ...      IdempotencyStore#check        SQL SELECT ...    HTTP -> risk-service  200 OK · 142ms
function callsSQLHTTPexceptionscode objectstimingrelationships

Observe broadly in production. Understand deeply in development. More information per run gives developers and coding agents better context before merge.

Find problems that only show up when the code runs.

Deep runtime evidence adds to what the code review shows. See performance problems, security and correctness issues, and unexpected changes while the code is still in development.

The traces give coding agents runtime knowledge they cannot obtain any other way, current at every commit.

Teams start free in their own repositories. AppMap Enterprise adds airgapped deployment, internal distribution, organization-registered installs, and telemetry into your internal observability systems for running Gold Trace programs at scale.

AppMap sequence diagram of one request, showing the full call path from HTTP to database.
Sequence diagram view. Fully interactive in your editor. Each trace captures the full request path from HTTP to database.
PERFORMANCE
  • Slow requests
  • N+1 queries
  • Repeated SQL
  • Expensive execution paths
SECURITY + CORRECTNESS
  • Auth gaps
  • Exceptions
  • Unsafe calls
  • Unexpected side effects
BEHAVIORAL CHANGE
  • API changes
  • Query changes
  • New dependencies
  • Unexpected drift in application behavior.

Know what your software does before you merge it.

Maps for developers. Runtime context for coding agents. Behavioral diffs before merge. Start free in VS Code or JetBrains.

NVIDIA Inception
GitHubfor Startups
MongoDB Partner
TechCrunch Finalist