backdoor is live · free · last verified 9 Nov 2026 · viability healthy
backdoor
Backdoor is a free, MIT-licensed Claude Code proxy that routes it to DeepSeek, Groq, Ollama, or OpenRouter instead of Anthropic.
If you're burning real money on Claude Code, this is the cheapest lever you'll pull this quarter: three commands, a three-line provider switch, and Python small enough to audit yourself. The trade is explicit. You're swapping Opus-class reasoning for price and inserting one maintainer's proxy into your inference path. Run it as a budget experiment with a security review attached, not as a drop-in upgrade.
Verdict generatedVerified 5d ago · liveness 70/100 · cite: rightaichoice.com/tools/backdoor
- Developers running Claude Code daily at high token volume who want the same harness on cheaper models
- Small teams trying to keep a large-scale code generation budget off the chopping block
- Privacy-conscious developers who need fully local inference via Ollama or LM Studio
- Students and solo builders leaning on the Groq or NVIDIA NIM free tiers
- Developers who need Claude Opus-class output quality on hard multi-step agentic tasks
- Non-technical users who want a GUI instead of a terminal wizard and .env edits
- Teams requiring official vendor support, an SLA, or production hardening
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Backdoor if your work genuinely needs Claude Opus-class output or you need a supported product with an SLA — it is an unsupported 18-commit MIT repo and your savings come directly out of model quality.
The proxy itself is free under MIT, but every routed token is billed by your chosen backend — DeepSeek, OpenRouter, and similar providers all charge per token, so the 99.95% savings headline still leaves a real bill.
Backdoor itself is free and MIT-licensed, so the pricing comparison is really between backends. At the README's own benchmark, 500M tokens costs $9,834 on Claude Opus 4.7 versus $5.34 on DeepSeek V3 Flash, with Groq and NVIDIA NIM free tiers covering smaller volumes at zero API cost. Set against that, a per-seat Claude Code or Copilot subscription is cheap for a single light user and expensive for a team running long agentic jobs.
In short
backdoor — Backdoor is a free, MIT-licensed Claude Code proxy that routes it to DeepSeek, Groq, Ollama, or OpenRouter instead of Anthropic. Best for Developers running Claude Code daily at high token volume who want the same harness on cheaper models, Small teams trying to keep a large-scale code generation budget off the chopping block, Privacy-conscious developers who need fully local inference via Ollama or LM Studio. Free to use.
What is backdoor?
backdoor — Backdoor is a free, MIT-licensed Claude Code proxy that routes it to DeepSeek, Groq, Ollama, or OpenRouter instead of Anthropic. Best for Developers running Claude Code daily at high token volume who want the same harness on cheaper models, Small teams trying to keep a large-scale code generation budget off the chopping block, Privacy-conscious developers who need fully local inference via Ollama or LM Studio. Free to use.
Backdoor is a small open-source proxy, written in Python and MIT-licensed, that lets you keep Claude Code as your coding agent while swapping the model underneath. It intercepts Claude Code's outbound API calls, translates the Anthropic request format into the OpenAI format, streams responses back, and preserves tool use, so agentic loops and the CLI experience carry on as normal. The backends documented in the README are DeepSeek, Groq, Ollama, OpenRouter, NVIDIA NIM, and LM Studio. Installation is three commands: git clone the repo, cd backdoor, then ./backdoor. That runs an animated terminal wizard that picks a provider, captures your key, writes the config, and launches a Claude Code session. Anyone who prefers editing files directly can copy .env.example, fill it in, and run ./run.sh. Switching providers later is a three-line change to PROVIDER_BASE_URL, PROVIDER_API_KEY, and PROVIDER_MODEL. The intended audience is developers who run Claude Code heavily and feel the token bill, small teams protecting a codegen budget, and anyone who wants the workflow on private or low-cost models. The repo's own benchmark pushed 500M tokens through the same workload at $9,834 on Claude Opus 4.7 versus $5.34 on DeepSeek V3 Flash. Groq and NVIDIA NIM come with free tiers, Ollama and LM Studio keep inference on your machine, and OpenRouter opens a large model catalog through one key. Against Claude Code Router, which routes from a Node base with a broader provider matrix, Backdoor's pitch is source you can read quickly and a wizard a non-DevOps developer can follow. It is not an Anthropic-sanctioned path, and it is not built for Opus-class reasoning on hard agentic tasks.
Key features
- Intercept Claude Code API calls and reroute them to any OpenAI-compatible provider
- Translate Anthropic request format into OpenAI format with streamed responses
- Preserve tool use so Claude Code agentic loops run unchanged
- Supported backends: DeepSeek, Groq, Ollama, OpenRouter, NVIDIA NIM, LM Studio
- Three-command install: git clone, cd backdoor, ./backdoor
- Animated terminal setup wizard picks your provider and writes config
- Three-line provider switch via base URL, API key, and model
- Local, private inference through Ollama or LM Studio with nothing leaving the machine
- Free-tier hosted backends: Groq and NVIDIA NIM
- OpenRouter single key unlocking a large model catalog including Gemini, Grok, Qwen, Command R
- Benchmark reference: 500M tokens cost $9,834 on Claude Opus 4.7 vs $5.34 on DeepSeek V3 Flash
- Cost calculator script (cost.py) for estimating provider spend
- Health check endpoint for monitoring proxy status
- Optional Telegram bot integration
- MIT-licensed Python source, documented as under 600 lines
Behind the verdict
We'd reach for Backdoor the moment a Claude Code habit starts showing up as a scary line on a card statement. The install is genuinely three commands, the wizard asks the right questions, and within a minute you are in the same Claude Code interface you already know. That is rare for routing tools, which usually expect you to be comfortable with Node configs and provider quirks. The economics are the draw, but read the benchmark carefully. The repo compared 500M tokens of the same workload and found $9,834 on Claude Opus 4.7 against $5.34 on DeepSeek V3 Flash. Those numbers are a demonstration of the spread, not a promise that DeepSeek will handle your hardest refactor as well as Opus. Budget with the cheap model, but keep a paid fallback for the jobs where reasoning quality shows. Where it bites: you are inserting an unaudited middleman between your source code and a third-party API. Backdoor itself is MIT-licensed and under 600 lines, which makes a security review tractable, but your legal or security team still has to sign off on proxying Claude Code traffic through a non-Anthropic endpoint. For teams with an SLA requirement or production hardening, this is the wrong shape of tool. Your closest alternative is Claude Code Router, which plays the same routing game from a Node base and carries a richer provider matrix. If you want breadth and are comfortable in JavaScript, look there. If you want something you can finish reading on a lunch break and hand to a colleague who flinches at config files, Backdoor wins. Adoption caveat: every time Claude Code ships an update, tool calls and streaming should be re-verified. That is the maintenance tax of any proxy and applies whether you pick this or a competitor. For local-only work, point it at Ollama or LM Studio and
Models under the hood
as of 2026-11-08
Limitations
- Backdoor is a free MIT-licensed local proxy, not an AI provider: it intercepts Claude Code's API calls and reroutes them to an OpenAI-compatible backend such as DeepSeek, Groq, Ollama, OpenRouter, NVIDIA NIM, or LM Studio.
- Claude Code must already be installed and configured, and output quality depends entirely on the chosen backend.
- It is a small personal open-source repository (301 stars, 32 forks) with no vendor support or SLA, and setup is a terminal wizard (git clone, cd backdoor, ./backdoor) rather than a GUI.
- Routing Claude Code to non-Anthropic providers may raise Anthropic terms-of-service questions.
as of 2026-10-09
Researching backdoor? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is backdoor still active?
backdoor is live · free · last verified 9 Nov 2026 · viability healthy.
Viability score
How well maintained and how widely used is backdoor? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: November 2026
How we score →What do users say about backdoor?
What people actually say about backdoor — is it worth it?
Our community sentiment research written Jul 26, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
How we verified this
Every listed fact about backdoor — pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 9 Nov 2026. Analysis is drafted with AI models and checked against our editorial standards before publishing; see how we verify.
Verification history
We have re-verified backdoor 17 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed summary, features, pricing tiers, description
- — changed pricing page
- — changed integrations, features, our verdict, description, summary, pricing tiers
- — changed pricing page
- — changed who it suits, description, our verdict, features, integrations, who should skip it, pricing tiers, summary
- — changed limitations, pricing page
Showing the 6 most recent of 17 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does backdoor cost in 2026?
backdoor: Free to use. Pricing last verified 9 Nov 2026.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where backdoor's pricing actually pencils out — and where peers do it cheaper.
Backdoor itself is free and MIT-licensed, so the pricing comparison is really between backends. At the README's own benchmark, 500M tokens costs $9,834 on Claude Opus 4.7 versus $5.34 on DeepSeek V3 Flash, with Groq and NVIDIA NIM free tiers covering smaller volumes at zero API cost. Set against that, a per-seat Claude Code or Copilot subscription is cheap for a single light user and expensive for a team running long agentic jobs.
Has backdoor's pricing changed?
No plan price for backdoor has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick backdoor?
Pick backdoor if you are Developers running Claude Code daily at high token volume who want the same harness on cheaper models, Small teams trying to keep a large-scale code generation budget off the chopping block, Privacy-conscious developers who need fully local inference via Ollama or LM Studio, Students and solo builders leaning on the Groq or NVIDIA NIM free tiers.
Look elsewhere if you are Developers who need Claude Opus-class output quality on hard multi-step agentic tasks, Non-technical users who want a GUI instead of a terminal wizard and .env edits, Teams requiring official vendor support, an SLA, or production hardening, Anyone whose legal or security team objects to routing Claude Code through a non-Anthropic proxy.
Skip it if Skip Backdoor if your work genuinely needs Claude Opus-class output or you need a supported product with an SLA — it is an unsupported 18-commit MIT repo and your savings come directly out of model quality.
GeneratedReal-world workflow fit
Concrete scenarios for the personas backdoor actually fits — and what changes day-one when you adopt it.
Clone the repo, run ./backdoor, let the wizard write a DeepSeek .env, and keep working in Claude Code exactly as before.
Outcome: Same agent interface, same tool use, but the monthly bill drops from Anthropic rates to roughly $0.01 per million tokens on DeepSeek V3 Flash.
Point Backdoor at a local Ollama instance so the proxy forwards to a model running on the same machine, with no external API in the path.
Outcome: Prompts and code never leave the machine, at the cost of running a smaller local model than a hosted frontier backend.
Run the same workload through Claude Code three times, editing the .env to swap DeepSeek, Groq, and an OpenRouter model between runs.
Outcome: A side-by-side comparison on actual work rather than benchmarks, with cost.py giving per-run numbers to pair with the quality results.
Use cases
- Route Claude Code through a local Ollama model for offline coding without sending code to external APIs
- Use Groq's fast inference with Claude Code to speed up code generation on supported models
- Replace Anthropic models with DeepSeek on cost-sensitive projects while keeping Claude Code's agent interface
- Benchmark Llama, Mistral, and DeepSeek against your real work using Claude Code as the harness
- Run Claude Code entirely on-premises by pointing Backdoor at a self-hosted model
- Leverage Groq and NVIDIA NIM free tiers to prototype without API costs
- Give a whole small team one provider switch instead of per-seat Anthropic token spend
Setup time & first value
How long it actually takes to get something useful out of backdoor — broken out by persona, not the marketing-page minute.
Solo developer: minutes — three commands (git clone, cd backdoor, ./backdoor), and the wizard writes the .env. You do need Claude Code installed and configured first, plus an API key for whichever backend you pick. Team or regulated environment: budget considerably longer for the security and legal review before anyone routes real code through it.
Switching to or from backdoor
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From plain Claude Code on Anthropic: install the proxy, keep Claude Code as-is, and let the wizard redirect the API calls to your chosen backend.
- →From Aider or Cline: keep Claude Code as the harness and put Backdoor in front of it rather than switching agents.
- →From a direct DeepSeek, Groq, or OpenRouter key in another tool: reuse the same key in Backdoor's .env instead of managing separate integrations.
- →From a hosted coding assistant: export your existing Claude Code settings and route through Backdoor for local or cheaper inference.
- ↗To plain Claude Code on Anthropic: delete the proxy config and restore Anthropic as the API endpoint — your Claude Code setup is untouched by Backdoor.
- ↗To Aider or Cline: point either tool directly at DeepSeek, Groq, or OpenRouter with your own key and drop the proxy layer.
- ↗To a hosted coding assistant subscription: cancel backend API keys and rely on the vendor's per-seat plan instead.
- ↗To a self-hosted coding stack: keep the Ollama model you were already running and consume it from a different client.
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “backdoor”, and we withheld 6: 6 could not be judged, because “backdoor” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about backdoor.
Official links
What are the best backdoor alternatives?
The closest alternatives to backdoor we list are NadirClaw, Not Diamond and OnAPI. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| NadirClaw | Freemium | Pricing not published | — |
| Not Diamond | Freemium | Pricing not published | — |
| OnAPI | Paid | Pricing not published | — |
Alternatives to backdoor
View all
NadirClaw
LLM router that picks the cheapest Claude, GPT or Gemini model per prompt and per Claude Code or Codex turn.
Not Diamond
Not Diamond routes each prompt to the best-fit LLM automatically — one chat inbox for GPT-4o, Claude, and Gemini.
Tools that pair well with backdoor
Common stack mates teams adopt alongside backdoor, with the specific reason each pairing earns its keep.
NadirClaw
LLM router that picks the cheapest Claude, GPT or Gemini model per prompt and per Claude Code or Codex turn.
Not Diamond
Not Diamond routes each prompt to the best-fit LLM automatically — one chat inbox for GPT-4o, Claude, and Gemini.
OnAPI
OnAPI routes GPT, Claude, Gemini, Sora and Veo through one multi-model AI API gateway and a single key
backdoor vs its closest rival
backdoor's closest head-to-head is with Spider Cloud — These two are not alternatives — they sit at different layers of a stack and a buyer would rarely shortlist one against the other. Backdoor is a free local proxy: if your Claude Code token bill is the problem, you install it, point it at DeepSeek, Groq, or Ollama, and you are done — no vendor contract, no support line, no production hardening. Spider Cloud sells the opposite thing: managed infrastructure for getting live web pages into an agent or RAG pipeline, with residential/ISP proxies, browser rendering, and a flat-rate Unlimited crawl tier. Buy Spider Cloud when the missing capability is web data at scale; install Backdoor when the missing capability is a cheaper inference path. If you must pick one today, pick the one that matches your actual bottleneck — cost per token, or access to rendered web content. Verdict generated
Head-to-head comparisons
Backdoor vs Spider Cloud
These two are not alternatives — they sit at different layers of a stack and a buyer would rarely shortlist one against the other. Backdoor is a free local proxy: if your Claude Code token bill is the problem, you install it, point it at DeepSeek, Groq, or Ollama, and you are done — no vendor contract, no support line, no production hardening. Spider Cloud sells the opposite thing: managed infrastructure for getting live web pages into an agent or RAG pipeline, with residential/ISP proxies, browser rendering, and a flat-rate Unlimited crawl tier. Buy Spider Cloud when the missing capability is web data at scale; install Backdoor when the missing capability is a cheaper inference path. If you must pick one today, pick the one that matches your actual bottleneck — cost per token, or access to rendered web content.
GeneratedBackdoor vs Temporal Ai
These aren't competitors — they sit on opposite sides of the stack. Temporal AI is an enterprise durability layer for AI agents and multi-step workflows; you buy it when a crashed worker or abandoned session must not erase state, and you pay for SSO, audit logs, SOC 2/HIPAA, and an SLA-backed path on Temporal Cloud. backdoor is a free, MIT-licensed local proxy that keeps Claude Code's UI and tool loop but swaps the billing path to DeepSeek, Groq, Ollama, OpenRouter, or NVIDIA NIM — a cost play, not a reliability play. A developer could theoretically run an agent harness on Temporal while routing model calls through a proxy, but nobody is choosing one instead of the other.
GeneratedBackdoor vs Audioeye
These two should not be on the same shortlist. Backdoor is a free, MIT-licensed local proxy that a developer installs in three commands to point an existing Claude Code setup at DeepSeek, Groq, Ollama, OpenRouter, NVIDIA NIM, or LM Studio — it's a cost-engineering tool for high-token-volume coding work. AudioEye is a paid managed accessibility service for organizations that need ADA/Section 508/WCAG 2.2 A/AA coverage, expert-written fixes, and VPAT/ACR paperwork. If you arrived here comparing them, you likely have two separate problems — an API bill and an accessibility compliance obligation — and they need two separate answers, not a head-to-head.
GeneratedFrequently asked questions
For makers of backdoor
- Get a Verified Listing for backdoor ($39): a written decision within 1 working day.
- Free listing for another tool: Free reviews are decided in order, within 3 weeks of the review starting.
Used backdoor? Help shape our editorial sentiment research.