AMP by CanyonTechs AI is an autonomous incident mitigation platform: it watches your production logs, diagnoses what broke, and opens a reviewable pull request with a fix — without anyone prompting it first. It launched on Product Hunt this week, running as version 2.4 in production, and connects to a repo the same way any GitHub or GitLab app would.
Short answer: AMP by CanyonTechs AI monitors production logs, diagnoses incidents by reading stack traces against your repo code, and opens a pull request on a feature branch — never pushing straight to production. It's free to start (2 repos, 5 auto-fixes/month), with a Professional plan at $25/month (10 repos, 100 auto-fixes/month) and custom Enterprise pricing above that.
I worked through AMP's own product page, its live pricing page, and its Product Hunt listing. The product is too new for independent track records to exist yet, so that's the best source available. When I tested the pricing numbers against what third-party aggregators had listed, a few figures didn't match CanyonTechs' own page. Everything below is pulled straight from the source, not aggregator copy.
What you'll need
A GitHub or GitLab repository AMP can open pull requests against, since that's how it delivers fixes — it never writes directly to a branch you didn't create for it. You'll also need a way to route production logs to it: CanyonTechs supports Vector, AWS CloudWatch, or its own lightweight log-agent, so check which one your stack already uses before signing up. A Jira project is optional but recommended, since AMP files a ticket with a full audit trail alongside every pull request it opens. None of this requires touching your production deploy pipeline — AMP reads logs and repo code, it doesn't get direct production access, which matters if a security team has to sign off before you connect anything. Accounts start free at canyontechs.ai, and CanyonTechs says the first fix can land within minutes of connecting a repo and pointing logs at it.
Step-by-step: setting up AMP by CanyonTechs AI
1. Create an account and connect a repository
Sign up and link a GitHub, GitLab, or local repository. Start with a repo you're comfortable having AMP open pull requests against — the free Starter tier caps you at 2 repositories anyway, so you can't spread this across your whole org on day one even if you wanted to.
2. Point a log source at AMP
Choose Vector, CloudWatch, or CanyonTechs' log-agent as your ingestion path. This is stage one of what CanyonTechs calls its three-stage pipeline — Ingest & Parse — where filters strip noise and pull out exception blocks before anything reaches the diagnosis step.
3. Start with one low-traffic service
Nothing in AMP's setup forces this order, but it's the sensible one. Confirm the ingest-to-fix loop works end to end on a service where a wrong pull request costs you nothing. Then wire in something that matters.
4. Let AMP diagnose the first incident
Once a real exception lands, AMP's diagnose stage reads the stack trace against your actual repository code, runs tests, and scores its own confidence before deciding whether to act. This is the step that separates AMP from a log alert — it's reasoning about your code, not just flagging an anomaly.
5. Review the pull request and the linked ticket
The remediate stage creates a feature branch, commits the fix, opens a PR, and files a Jira or GitLab ticket with the reasoning attached. CanyonTechs is explicit that this is a hard boundary, not a setting you can turn off: AMP opens a PR on a feature branch and stops there, so a human still merges it.
6. Watch your fix budget before scaling up
Auto-fixes are metered per month, not per week — 5 on Starter, 100 on Professional — so check usage before connecting more repositories than your tier's fix budget can realistically cover.
Example prompts you can copy
AMP doesn't take an open-ended chat prompt — it works off logs and repo code, not a conversation — so "prompting" it mostly means setting up the inputs that make its diagnosis step accurate. A few things worth copying into your own rollout:
- A seed exception to test the pipeline first: throw a deliberate, low-stakes error (a null reference in a throwaway endpoint works well) at your log source before connecting anything real, so you can watch the ingest-diagnose-remediate loop complete once without risking a real service.
- A PR template note for AMP's pull requests: "This PR was opened automatically. Confirm the linked test run passed and that the fix addresses the root cause in the stack trace, not just the symptom, before merging."
- A ticket-triage label convention: tag AMP-filed Jira or GitLab tickets with something like
amp-autoso your team can filter them separately from human-filed bugs during standup. - A weekly fix-rate check-in: "How many of AMP's pull requests merged as-is this week versus needed changes?" — a simple number to track for the first month, since that ratio tells you more about whether to trust it unattended than the vendor's published 82% figure does for your specific codebase.
Common mistakes to avoid
The first mistake is reading "5 auto-fixes" or "100 auto-fixes" as a weekly allowance — it's monthly, and a noisy service can burn through the Starter tier's 5 fixes in a single bad day. Second, assuming AMP covers your whole stack: it's certified today for Java, Python, JavaScript, TypeScript, and Node.js, so a Go or Ruby service won't get the same diagnosis quality, if it's covered at all. Third, connecting a critical production repository before the ingest-to-fix loop has run once on something disposable — the product is brand new, at version 2.4 as of this writing, and there's no long usage history to lean on yet. Fourth, skipping the Jira or GitLab connection and then wondering why fixes show up as PRs with no ticket trail — that link is what gives you the audit history CanyonTechs advertises. Fifth, treating the Product Hunt launch promo code (PH3MOFREE, three months free on a monthly plan) as a standing offer — CanyonTechs states it's valid for the launch window only, so confirm it's still live before budgeting around it.
AMP by CanyonTechs AI pricing
| Tier | Price | Repos | Auto-fixes | Also includes |
|---|---|---|---|---|
| Starter | $0/month | Up to 2 | 5/month | Basic LLM access |
| Professional | $25/month | Up to 10 | 100/month | 99.9% uptime SLA, priority ticket support, advanced code analysis, CI/CD integration |
| Enterprise | Custom | Unlimited, multi-org | Unlimited | 99.99% uptime SLA, indefinite audit logs, on-premise deployment, custom LLM support, dedicated solutions architect |
I checked every number in that table directly against CanyonTechs’ own pricing page on August 13, 2026. A same-week Product Hunt launch is exactly the kind of thing that moves fast, so check the live page before you commit a team to a specific tier. CanyonTechs also reports an 82% autonomous fix rate and a median time-to-resolution under 4 minutes across its own usage — figures worth treating as vendor-reported until independent reviews catch up, since the Product Hunt listing is only days old.
Tools that make this easier
AMP is one piece of an incident-response workflow, not the whole thing, and it's worth knowing what it pairs with. If your team wants a second opinion on whether AI-generated fixes hold up under real load before trusting them unattended, my honest review of AI programming tools is a useful gut check, and the broader AI coding assistant guide covers the tools that write the code AMP is patching. Since AMP's whole pitch starts with reading production logs, pairing it with a dedicated AI observability setup gives you better signal before an exception even reaches AMP's ingest stage. Teams running an AI gateway in front of their services should also see how ngrok’s AI gateway fits into that same production layer. On the ticket side, AMP files into Jira or GitLab directly, but if you want a dedicated AI layer for triage and routing, TicketDesk AI covers that angle. And because paid tiers here scale with repos and fix volume, not seats, managing AI coding costs at scale is worth reading before you connect more than a couple of services. For a wider view of what else is worth trying in this category, my best AI tool for code roundup is the place to start.
My take
The pitch — logs in, diagnosis and a reviewable PR out, with a hard rule against touching production directly — solves a real problem for teams tired of being paged for the same class of bug repeatedly. The human-in-the-loop design is the right call for a tool this new: every fix lands as a pull request on a feature branch, not a live change, so the worst case is a bad PR someone has to close, not an outage AMP caused itself. What I can't verify yet, because the product launched this week, is how the 82% fix-rate figure holds up on a real, messy codebase over months rather than a vendor's own benchmark. Start on the free Starter tier against one low-traffic service, watch a handful of its pull requests before trusting it further, and don't let the monthly fix cap surprise you if you're evaluating this on a noisy service.
Last updated: August 13, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Frequently Asked Questions
Is AMP by CanyonTechs AI free?
Yes, to start. The Starter tier is $0/month with up to 2 connected repositories and 5 auto-fixes per month, using basic LLM access. Paid plans begin at $25/month once you need more repos or a higher fix budget.
How long does it take to set up AMP by CanyonTechs AI?
Connecting a repo and pointing a log source at AMP takes a few minutes, and CanyonTechs says the first fix can land shortly after. Getting real confidence in it takes longer — plan on watching its first several pull requests closely before letting it run against anything critical unattended.
What is the easiest way to start with AMP by CanyonTechs AI?
Connect one low-traffic repository, wire up a single log source, and let it complete the ingest-diagnose-remediate loop once on something disposable before adding a second service. The free Starter tier's 2-repo cap makes this the natural on-ramp anyway.
Does AMP push fixes directly to production?
No. AMP opens a pull request on a feature branch and stops there — it doesn't have direct production access, and every fix requires a human to review and merge it before it ships.
What languages does AMP by CanyonTechs AI support?
As of version 2.4, it's certified for Java, Python, JavaScript, TypeScript, and Node.js. Other languages aren't listed as supported yet, so confirm current coverage on CanyonTechs’ site before connecting a service written in something else.