Works with 13+ AI coding agents — Claude Code, Codex, Gemini CLI, Grok

One lens on every
AI coding agent

Your team runs Claude Code, Codex, Gemini CLI, OpenCode, Grok and whatever ships next week. AgentLens pulls all of them into one dashboard — usage, spend, session history and real performance signals, per developer, per project, per team.

npx agentlens connect --all

Agent spend
$4,182
+12.4%
Sessions
9,741
+7.1%
Tokens
412M
+18.9%
Cost / merged PR
$3.10
−9.6%

Tokens by agent

8 weeks
  • Claude Code184M
  • Codex119M
  • Gemini CLI81M
  • OpenCode48M

Share of spend

  • Claude Code44%
  • Codex26%
  • Gemini CLI17%
  • OpenCode13%

Sessions by day

stacked by agent
  • Claude Code
  • Codex
  • Gemini CLI
  • OpenCode
76Mon
90Tue
97Wed
103Thu
121Fri
32Sat
25Sun

Spend by project

this month
payments-api$1,482
web-app$1,094
mobile$812
infra$524
design-system$270

Activity heatmap

day × hour band
Mon
Tue
Wed
Thu
Fri
Sat
Sun
LessMore

Tool-call success

last 7 days
Claude Code
94%
Codex
88%
Gemini CLI
81%
OpenCode
73%
Live agent activity
14:02claude-coderefactor webhook retries$0.41
14:01codexfix hydration mismatch$0.12
13:58gemini-cliwrite terraform module$0.08
13:55opencodeadd offline cache layer$0.23
Sample AgentLens overview: spend, sessions, tokens and cost per merged PR; tokens by agent over eight weeks; share of spend by agent; sessions per weekday stacked by agent; spend by project; an activity heatmap by day and hour band; tool-call success rate per agent; and a live activity feed.

Vendor-neutral by design — connect the agents you already use

CCClaude Code
CXCodex CLI
GEGemini CLI
OCOpenCode
GRGrok CLI
QWQwen Code
GHGitHub Copilot
CUCursor
AIAider
CLCline
WSWindsurf
COContinue
AMAmp
{ }Custom agent
4.7agents

run by the average engineering org we talk to

38%

of AI spend teams can't attribute to a project

6 min

median time to first chart after install

100%

of collectors are read-only and self-hostable

Platform

Everything you can't see today

AI coding tools are now a real line item and a real part of your delivery pipeline. AgentLens gives engineering leaders the same visibility they already expect from CI, cloud spend and incident tooling.

One dashboard, every agent

Sessions, prompts, tool calls and token counts from all your agents, normalised into a single schema. Stop comparing four vendor consoles that disagree with each other.

Real cost attribution

Spend rolled up by developer, project, team and agent — with per-model pricing applied automatically.

Agent performance scorecards

Tool-call success rate, retries, edits reverted, session length and time-to-done. See which agent is actually best for which repo.

Searchable history

Full-text search across every prompt, response and diff your agents produced. Filter by user, project, agent or date range.

Outcome linking

Tie agent sessions to the commits and PRs they produced, so you can report cost per merged PR instead of cost per token.

Budgets & alerts

Per-team monthly caps with Slack, email and webhook alerts before you blow through them — not three weeks later on an invoice.

Teams & projects model

Group repos into projects, projects into teams, and give each team its own budget, retention and access rules.

Open ingest API

Anything not on the connector list can push events over REST or OpenTelemetry. In-house agents included.

Your data, your perimeter

Self-host the whole stack or run it in our cloud. Redact prompt bodies at the edge and keep metrics only, if that is your policy.

How it works

Connected in about four minutes

No proxy in front of your model calls, no wrapper CLI your developers have to remember. AgentLens reads the data your agents already write.

01

Install the collector

One command on each machine, or a single deployment in CI for shared runners. Node 18+, no daemon, no IDE plugin.

$ npx agentlens connect --all
 
✔ detected claude-code ~/.claude
✔ detected codex ~/.codex
✔ detected gemini-cli ~/.gemini
✔ detected opencode ~/.local/share/opencode
→ linked to team platform-eng
02

Map projects & teams

Repos are grouped into projects, projects into teams. Budgets, retention and access all hang off that tree.

# agentlens.yml
team: platform-eng
projects:
payments-api:
repos: [org/payments-api]
budget_usd: 900
web-app:
repos: [org/web, org/design-system]
redact_prompts: false
03

Watch it land

Metrics stream in within seconds. Historical logs already on disk get backfilled on first sync, so day one is not an empty chart.

$ agentlens status
 
agents 6 connected
projects 14 mapped
backfill 92 days imported
spend MTD $4,182 / $6,000
alerts 1 (mobile 84% of budget)
Integrations

Bring your own agents

Every connector is read-only and runs on your side of the wire. If a tool writes a session log or exposes a usage API, AgentLens can read it — and the open ingest API covers everything else.

Request a connector
CC

Claude Code

Live

Anthropic

Local session logs + OTel hooks

CX

Codex CLI

Live

OpenAI

Session rollouts + API usage

GE

Gemini CLI

Live

Google

Telemetry export + local history

OC

OpenCode

Live

SST

Storage dir + server events

GR

Grok CLI

Beta

xAI

API usage + local transcripts

QW

Qwen Code

Beta

Alibaba

Local history + DashScope usage

GH

GitHub Copilot

Live

GitHub

Copilot metrics API + CLI logs

CU

Cursor

Beta

Anysphere

Admin usage API

AI

Aider

Live

Open source

Chat history + analytics file

CL

Cline

Beta

Open source

Task store + provider usage

WS

Windsurf

Coming soon

Codeium

Teams analytics API

CO

Continue

Coming soon

Open source

Dev data events

AM

Amp

Coming soon

Sourcegraph

Thread export API

{ }

Custom agent

Live

Your stack

Ingest REST API + OpenTelemetry

Pricing

Priced on teams, projects and connected agents

Not per seat, not per token. You pay for the shape of your organisation — how many teams you roll up, how many projects you track, and how many AI tools you plug in. Unlimited sessions and unlimited tokens on every plan.

MonthlyYearlySave 20%

Solo

For one developer keeping an eye on their own agents.

$0forever
1
Teams
3
Projects
2
AI tools connected
  • 1 seats · 14 days retention
  • Unified session & cost dashboard
  • Prompt and conversation search
  • Local-first sync CLI
  • Community support
Start free

Pro

For small squads shipping with several agents at once.

$23/ mo

billed annually

2
Teams
15
Projects
5
AI tools connected
  • 5 seats · 90 days retention
  • Everything in Solo
  • Per-project cost & token breakdowns
  • Agent performance scorecards
  • Budget alerts via email & Slack
  • CSV / JSON export
Start 14-day trial
Most popular

Business

For engineering orgs standardising on AI tooling.

$79/ mo

billed annually

10
Teams
100
Projects
12
AI tools connected
  • 50 seats · 1 year retention
  • Everything in Pro
  • Multi-team rollups & cross-team compare
  • SSO (Google / Okta) + role-based access
  • Ingest API, webhooks & OpenTelemetry
  • Chargeback reports per team
  • Priority support
Start 14-day trial

Enterprise

For regulated teams that need it inside their own perimeter.

Custom
Unlimited
Teams
Unlimited
Projects
Unlimited
AI tools connected
  • Unlimited seats · Unlimited retention
  • Everything in Business
  • Self-hosted or private cloud
  • SAML & SCIM provisioning
  • Audit logs & data residency
  • Custom agent connectors
  • 99.9% SLA + named engineer
Talk to us

Size your plan

Drag the three axes we price on and we'll point at the cheapest plan that covers you.

3
25
6
Recommended
Business$79 / mo

Covers up to 10 teams, 100 projects and 12 connected AI tools.

Need a little more? Add on:
  • Extra team+$15 per team / mo
  • Extra 10 projects+$10 per pack / mo
  • Extra agent connection+$9 per agent / mo
  • Extra 10 seats+$25 per pack / mo

All plans include unlimited sessions, unlimited tokens tracked and the open ingest API. Self-hosting is free forever for Solo.

Security

Built for teams that can't ship their code to a dashboard

AgentLens sits beside your agents, not in front of them. There is no proxy, no model key held on our side, and nothing in the path that can break a developer's session. If our service is down, your agents keep working and the collector backfills later.

No model keys storedRead-only connectorsOffline bufferingOpen source collector

Self-host in one container

Docker image, Postgres or SQLite, no outbound calls required. Your agent transcripts never leave your network.

Redaction at the collector

Strip prompt and response bodies before they are sent, keeping only metrics and metadata. Secret patterns are masked by default.

SSO, RBAC, scoped tokens

Google and Okta SSO, SAML and SCIM on Enterprise, per-team roles, and revocable per-machine collector tokens.

Audit-ready

Immutable audit log of every view, export and settings change, plus configurable retention and data residency.

FAQ

Questions engineering leads ask us

What counts as a “connected AI tool”?

One agent product per workspace — Claude Code, Codex CLI, Gemini CLI, OpenCode, Grok, Qwen Code and so on. Connecting the same agent on fifty developer machines still counts as one, and every model that agent calls is included.

Do you sit between my agent and the model provider?

No. There is no proxy and we never hold your provider API keys. The collector reads the session logs and usage endpoints your agents already write, so nothing in your inference path changes.

Can I track an in-house or unsupported agent?

Yes. Push events to the ingest REST API or point your existing OpenTelemetry exporter at AgentLens. Custom agents get the same dashboards, cost attribution and scorecards as the built-in connectors.

How is cost calculated for agents that don't report it?

We apply a maintained per-model price table to the token counts in the session log, and use the provider's own reported spend wherever a usage API exposes it. You can override rates for negotiated or self-hosted models.

Is my source code or prompt content stored?

Only if you want it. Prompt and diff bodies are optional — turn on redaction at the collector and AgentLens keeps metrics and metadata only. On Enterprise you can run the whole stack inside your own perimeter.

What happens if I outgrow a plan mid-month?

Nothing breaks. You keep collecting, we flag the overage in-app, and you either upgrade or attach an add-on for the extra team, project pack or agent connection. We never drop data because of a limit.

Do you have a free tier for open source?

Yes — public repositories get Business-level limits at no cost. Email us with the org and we'll switch it on.

Find out what your agents actually cost

Connect your first two AI tools in four minutes. Free forever for solo developers, and your historical logs get backfilled on the first sync.