Mission control for your coding agents
Take control of your coding agents’ SPEND
75%+ off your agent bill · 4× longer sessions. Plus insight into what your agents actually did — every decision, traced. Every agent.
- SPEND — Cut your coding-agent bill 75%+ — and run 4× longer sessions — on whatever agent you already use.
- CONSISTENCY — Every developer's agent writes code your way — your conventions, enforced automatically.
- VISIBILITY — One view across every agent and repo — Claude Code, Cursor, Copilot — not a dashboard per tool.
- INSIGHT — See what the AI actually did — the decisions it made, and whether your engineers are still steering.
Benchmarked — reproduce it on your own repo.
The problem
AI writes more of your code every week. Your spend, your standards, your visibility, your insight — none of them kept up.
The bill is unpredictable
Coding-agent spend swings month to month and nobody can explain it — quotas burn out mid-task, costs climb, and there's no line item that says why.
The code is drifting
Every developer's agent writes the code a little differently. Conventions erode one merged PR at a time, and your codebase slowly stops looking like one team wrote it.
There's no single view
Claude Code here, Cursor there, Copilot somewhere else — every tool keeps its own dashboard, and not one of them shows you the team.
The decisions disappear
You're paying per token but can't tell what it bought — which calls the AI made, where it went off-script, or whether your engineers are still steering.
How it works
One install. It does the rest from inside your agents.
- 1
Install once
One npm install puts unerr on your machine. One `unerr install` wires it into Claude Code, Cursor, Copilot — or any agent you run, per repo. Restart your IDE and the next prompt already knows your repo. Your source code stays on your machine.
- 2
It works inside the agent
unerr gives your agent a live map of your code and your team's conventions, so it stops re-reading everything and stops drifting from house style.
- 3
You get the picture
One view of spend, consistency, and the decisions your agents made — across every tool and repo. Watch it live on the local dashboard from the moment you install.
$ npm install -g @unerr-ai/unerr
$ cd ~/your-project
$ unerr install claude-code
✓ MCP config, skills & hooks written
✓ Live map of your repo ready — <5ms lookups
✓ Dashboard live at localhost:9847
Restart your IDE — the next prompt already knows your repo.Spend
75%+ off your bill. 4× longer sessions. Every agent.
Don't take the number — reproduce it. Your token bill has a cheap side and an expensive side. Point tools only trim the cheap side. unerr trims both — one bar per tool below, split the way your bill is.
Input · measured
Output & beyond · 5–6× the input cost
unerr
Code-graph tools
CLI text compressors
Fill = tokens saved in each zone — bigger is better · input: −navigation% / −read%, measured (real tokenizer, fidelity-gated, best result per category) · full write-up
The measured number is input-only. Your bill isn’t.
Those percentages are input compression — the cheapest tokens you buy. The point tools stop there. unerr keeps working where tokens cost 5–6× more: fewer round trips, flat context, cache-safe prompts, leaner output.
| Token lever | Code-graph tools | CLI text compressors | unerr |
|---|---|---|---|
| Input compressionthe measured slice · the cheapest tokens on the bill | |||
| Code exploration reduction | −43% / −81% | — | −86% / −90% |
| Command-output compression | — | core focus | ~89% avg |
| Beyond inputoutput tokens cost 5–6× more — no category plays here | |||
| LLM round-trip reduction | — | — | One call, not five |
| Context management | — | — | Digests, not dumps |
| Prompt & cache management | — | — | Cache-safe by design |
| Model-output reduction | — | — | Lean results, no replay |
Each category is genuinely good at its slice. unerr covers both slices — and the expensive levers beyond them. Want your own numbers? estimate your agent’s token waste
How it lands on the agent you already pay for
| Agent | Type | What caps you | How the saving lands |
|---|---|---|---|
| Claude Code | Terminal | ~45 msgs/5h (Pro); weekly caps even on Max | Leaner messages → ~3× longer sessions, quota stretches |
| Cursor | IDE | $20 credits/mo; Claude burns ~2.4× faster | Cleaner requests → credits go ~2× further |
| OpenAI Codex | Agent | 30–1,500 msgs/5h by plan | Compressed command output → more iterations per window |
| Windsurf | IDE | 500 credits/mo (Pro) | Fewer tokens per prompt → credits last ~2× longer |
| Gemini CLI | Terminal | 1,000 req/day free, then pay-per-token | ~70% less on the token bill / rate-limit headroom |
| GitHub Copilot | IDE | 50–1,500 premium req/mo | Leaner terminal output → more context per premium req |
Per-agent translations are illustrative against current plan limits — list prices change; the measured token reduction is the durable claim.
Insight
See the decisions — and who's steering.
What did the AI actually do with the tokens you paid for? unerr keeps the decision trail — the choices your agents made, the reasoning quality of the code, and whether your engineers are still in charge. Team-level, never per-person surveillance.
Who's steering — your team's profile
Built from how your team prompts and interacts with the AI — how it directs, corrects, reviews, and hands off.
Where the tokens went — and what landed
Spend split by what the AI was actually doing, next to how much of the work was planned upfront and landed clean.
Planned upfront, landed first try
72%
Changed again within two weeks
18%
Illustrative preview — the dashboard reports these live, per repo and per team. Always team-level.
Consistency
Every agent writes code your way.
unerr learns the patterns your codebase already follows — naming, structure, idioms — and applies them to every agent as the code is written. No style guide to hand-maintain, no drift between devs.
Four catches, one edit
first attempt → house pattern
function getUsrNm(u) {function getUserName(user) {learned from your repo — no style guide written
const data = fetch(url)const data = await safeFetch(url)the house pattern, applied every time
query(`SELECT … ${input}`)query(sql, [input])input bound — injection path closed
color: "#8B5CF6"color: tokens.accentdesign tokens, never raw values
Caught as the code is written — not at CI, not in review.
Your house rulebook — mined, not written
unerr digs most of the rulebook out of your repo on its own; your team tops it up where it matters. Every rule lands on every agent.
Conventions
40 rules
Guardrails
21 rules
Security rules
13 rules
Error handling
14 rules
247catches enforced this month — drift stopped before it reached review.
Illustrative preview. Ten developers, ten agents — the codebase keeps reading like one team wrote it, and review time goes back to architecture instead of naming. Why agents drift in the first place: they forget everything between sessions.
Visibility
One view across every agent and repo.
Claude Code, Cursor, Copilot — not a dashboard per tool. One place that shows what your agents are doing across every repo, every developer, every session. Local-first: the view comes to you, your code goes nowhere.
Every agent you run, lined up
Most teams run two or three agents side by side — each one's own analytics stop at its own walls. This is the one place they line up.
Claude Code
128 sessions · 9 drift catches
Cursor
87 sessions · 14 drift catches
GitHub Copilot
41 sessions · 6 drift catches
Gemini CLI
22 sessions · 3 drift catches
One session, traced end to end
The code-intelligence and reasoning-trace pages follow each session across agents — what was mapped, what was applied, what was caught — on the local dashboard.
- 14:02Cursor
payments-api — task opened, 12 files mapped before the first read
- 14:03Cursor
2 conventions applied · 1 breaking change flagged before the edit
- 14:11Claude Code
same thread picked up in the terminal — context carried over
- 14:19Claude Code
session closed — decision trail kept, every choice on record
The thread survives the tool switch — Cursor to terminal, nothing relearned.
Illustrative preview — the local dashboard reports these live, per repo and per team. Always team-level, never per-person surveillance.
One engine underneath. Not a patchwork of competing, inconsistent plugins.
Read it bottom-up. unerr turns your code into a live map, layers intelligence, memory, tools and context on top, and hands the whole thing to every agent you run. The four sections above aren't four products — they're four reads of the same engine.
Your coding agents
all pulling from the same engine
the engine
runs on your machine · answers in <5 ms
Code graph
a live map of every file, function and call — rebuilt as you work
Your code
stays on your machine — mapped, never moved
Bottom to top: your code, one engine, every agent, four outcomes. The colored blocks jump back to their sections.
Local-first
Runs on your machine, next to your agent.
Your code stays yours
Source never leaves your machine.
Works with what you pay for
Claude Code, Cursor, Copilot & more.
No rip-and-replace
One install into your existing setup.
Pricing
One flat rate. It pays for itself.
No credits, no usage-based billing — the savings on your agent bill cover the seat.
Free
Your agent, one repo, on your machine. No credit card.
- No code leaves your system
- Full local core — every agent
- 1 repo · unlimited memory storage
Pro
7-day free trial, then $25/mo
Everything that compounds — for one developer.
- No code leaves your system
- Unlimited repos
- Analytics processed — savings & usage insight
- Full dashboard history
No credit card required · Cancel anytime
Teams
For engineering teams that run on coding agents.
- 75%+ avg bill cut, every agent
- Cross-repo intelligence between repos
- Conventions enforced team-wide
- Patterns & skills insight
- Flat rate — no credits, no surprises
Enterprise
Self-hosted, security review, custom terms.
- Everything in Teams
- One-time LLM application cost analysis
- Self-hosted deployment
- Security & compliance review
- Dedicated support
The full comparison matrix and what each gate means — compare all plans
Take control of your coding agents.
One install. 75%+ off the bill, code that stays consistent, and a clear picture of what the AI actually did.