Rudel
Open-source session analytics for Claude Code and Codex, self-hostable and free
Rudel is an open-source analytics layer for AI coding agents. It ingests full Claude Code and Codex session data — tokens, duration, model usage, git context and transcripts — so engineering leaders can see how their teams actually use coding agents rather than guessing from seat counts and anecdote.
Rudel is an MIT-licensed, open-source analytics dashboard for AI coding-agent sessions, covering Claude Code and Codex. It ingests full session data rather than summary counters: session identifiers and timestamps, user and organisation, project path and package name, git context including branch, commit SHA and remote, complete session transcripts, and sub-agent usage. From that it reports token consumption, session duration, activity patterns and model usage, giving engineering leaders a measured view of coding-agent adoption instead of seat counts and anecdote. The stack is Bun as runtime, a Turbo monorepo, Biome for linting and ClickHouse for storage, with Docker and Fly.io deployment paths documented for self-hosting; the hosted service runs at app.rudel.ai. Setup is three commands — npm install -g rudel, rudel login, then rudel enable to auto-upload sessions. The project publishes its own findings from a corpus of 1,573 real Claude Code sessions and over 15 million tokens, and those numbers are the strongest argument for the category: skills were invoked in only 4 percent of sessions, 26 percent of sessions were abandoned with most abandoned inside the first 60 seconds, and success rates varied sharply by task type. It launched as a Show HN on 12 March 2026, reaching 144 points, and on Product Hunt on 9 April 2026, where the product was subsequently rebranded to nua by early May 2026 while the site and repository still carry the Rudel name. Privacy is the live question: the tool redacts known secret patterns before upload but its own documentation warns against treating that filter as a substitute for keeping secrets out of sessions in the first place, which is why self-hosting via Docker is the answer the maintainers give to enterprise objections.
An engineering leader who has bought Claude Code or Codex seats at scale and has no evidence of whether they are used, abandoned or wasted.
Measured session-level evidence of coding-agent adoption — abandonment rates, skill usage, token spend by project — from a tool you can run entirely inside your own infrastructure.
At a Glance
- Category
- Developer Tools
- Pricing
- Free, Freemium
- Target Market
- CTOs, VPs of Engineering, Engineering Managers, Platform Engineers
- Deployment
- Open-source, Self-hosted, Cloud-first
Key Features
- ✓Full session transcript ingestion
Captures complete Claude Code and Codex sessions rather than summary counters, so patterns can be reconstructed after the fact.
- ✓Git and project context capture
Records project path, package name, branch, commit SHA and remote, letting agent spend be attributed to specific repositories.
- ✓Token and duration analytics
Reports token consumption, session length, activity patterns and model usage across users and organisations over time.
- ✓Sub-agent usage tracking
Surfaces when and how sub-agents are invoked within a session, which most seat-based reporting cannot see at all.
- ✓Docker and Fly.io self-hosting
Documented self-hosting path with ClickHouse storage, so transcripts never leave your infrastructure if policy forbids it.
- ✓Secret redaction before upload
Filters known secret patterns from sessions prior to upload, with explicit documentation that it is not a guarantee.
- ✓Three-command install
npm install -g rudel, rudel login and rudel enable turn on automatic session upload without changing developer workflow.
Capabilities
Use Cases
- •Justifying or cutting coding-agent seat spend
Show which teams genuinely use their Claude Code seats and which have abandoned them after the first week.
- •Diagnosing agent session abandonment
Identify the task types and prompts where developers give up within 60 seconds and fix the underlying friction.
- •Auditing skill and sub-agent adoption
Determine whether the custom skills your platform team built are ever invoked in real developer sessions.
- •Chargeback of agent token spend by repository
Attribute token consumption to specific projects using captured git remote and branch metadata for internal cost allocation.
- •Privacy-constrained internal rollout
Run the full analytics stack on Docker inside your own network so no transcript ever reaches a third-party service.
Ideal For
Best For
- ✓Measuring real Claude Code and Codex adoption across a team instead of inferring it from seat licences
- ✓Finding where agent sessions are abandoned, given the project's own finding that 26 percent are dropped and most within 60 seconds
- ✓Auditing which agent skills are actually invoked — the project measured just 4 percent skill usage across 1,573 sessions
- ✓Attributing agent token spend to specific projects and git repositories using captured branch, SHA and remote context
- ✓Privacy-constrained organisations that need the analytics but must keep transcripts inside their own perimeter via Docker self-hosting
Not Ideal For
- ✗Teams using Cursor, GitHub Copilot, Gemini CLI or any agent other than Claude Code and Codex — those are simply not supported
- ✗Organisations unwilling to upload verbatim session transcripts and unable to run a self-hosted ClickHouse deployment, which is the only alternative
- ✗Anyone whose need is met by Claude Code's built-in /insights command, which HN commenters noted overlaps part of this ground
- ✗Small teams where the operational cost of a ClickHouse-backed analytics stack outweighs what the numbers would change
Deployment
Market Analysis
Pros
- ✓Fully free and MIT licensed with a documented Docker self-hosting path, so there is no licence barrier to evaluating it
- ✓Captures a genuinely under-instrumented layer — session transcripts and sub-agent usage that seat-based vendor reporting does not expose
- ✓The published corpus of 1,573 sessions produced findings worth the tool on their own: 4 percent skill usage, 26 percent session abandonment
- ✓Reached 144 points on Hacker News in March 2026, so the design has had real practitioner scrutiny
- ✓Active project at 288 stars, 16 forks and 446 commits, on a modern Bun and ClickHouse stack
Cons
- ✗HN commenters balked at uploading Claude Code sessions verbatim to a third party without clear security documentation — the maintainers' answer is self-hosting, which shifts the burden to you
- ✗The project's own docs warn that its secret-redaction filter is not a substitute for keeping secrets out of sessions, so leakage risk is real
- ✗Commenters could not tell exactly what is sent to the LLM as context or how output is processed, calling the analysis a black box
- ✗Claude Code's built-in /insights command already covers part of the same ground, raising a genuine question of marginal value
- ✗Some reviewers found the architecture overengineered for the size of the dataset it was demonstrating on
- ✗Skeptics questioned whether the 1,573-session dataset exists as described, since no public dataset was released
- ✗Only Claude Code and Codex are supported — no Cursor, Copilot or Gemini CLI — and the product was rebranded to nua on Product Hunt while the site and repo still say Rudel, which makes it hard to track
Pricing
Open source / self-hosted
$0
- ✓MIT-licensed source on GitHub
- ✓Docker and Fly.io deployment paths
- ✓ClickHouse-backed storage under your control
- ✓All analytics features, no functional gating
Hosted (app.rudel.ai)
$0
- ✓Managed instance, no infrastructure to run
- ✓Three-command install via npm
- ✓Secret-pattern redaction before upload
- ✓Team and organisation-level reporting
The project describes itself as fully free and self-hostable, with no published paid tier for either the hosted app.rudel.ai instance or the MIT-licensed self-hosted build. That means the real cost is operational rather than licensed: self-hosting requires running and maintaining a ClickHouse deployment, and no commercial support agreement or SLA is offered, which is the trade a regulated buyer has to weigh against uploading verbatim transcripts to a hosted service.
Security & Compliance
Connect
Sources
This page was written from 4 sources, 3 on domains other than rudel.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Confident AI
Hosted LLM evaluation, observability and red teaming built on the open-source DeepEval framework
Respan
Observability, evals and an LLM gateway for AI agents in one control plane
Meta Muse Code
Meta's terminal coding agent for large repositories, with persistent background agents and the most aggressive token pricing in the category
Niteshift
The full-stack cloud for coding agents — real environments, verified pull requests