Argus is free. You pay for tokens.
No seats, no tiers, no gated features. Argus is open source under AGPL-3.0 and runs on your infrastructure, so you pay for the services it calls and the servers it runs on, not for Argus.
- Argus
- $0. Open source under AGPL-3.0, with no paid tier.
- Model tokens
- Your LLM provider bills your own key for each review. The estimator below prices it.
- Embeddings
- Your embeddings provider, for the finding and memory text it embeds.
- Sign-in
- Clerk, at its own rates, for the dashboard.
- Servers
- Wherever you run the Go server, the dashboard and Postgres.
What you run
Argus is self-hosted only; there is no public hosted service. You run the Go server and the dashboard, and your code, diffs and team memory stay in your Postgres.
- Postgres with pgvector 0.8.2 or later (the compose file ships Postgres 16 with pgvector)
- Your own GitHub App, with the permissions and events listed in the guide
- A Clerk instance for dashboard sign-in
- An API key for any of 11 LLM providers, and an
ENCRYPTION_KEYto store it - An embeddings endpoint for memory search (
EMBEDDINGS_API_KEY, or a local endpoint) docker compose upstarts Postgres, runs migrations and starts the server; Fly.io deploys work too
Nothing is held back
There is no paid tier. Each feature is a setting, and most are on by default; Deep Review is off by default because it costs more tokens.
- No repo or monthly review quota
- Deep Review: four specialist passes per file, opt-in
- Judge-scored findings, per-PR review contract
- Team memory from fixes, dismissals and conventions
- Earlier-finding re-checks and code-graph blast radius
- Graph-grounded Mermaid diagrams in the PR description
- Your own LLM key, with a model per pipeline stage
- Per-stage token counts on each posted review, with cost when the price is known
What you'll spend on tokens
Your model provider's bill is the cost that grows with use. Pick a model and a monthly review volume to estimate it at OpenRouter prices, then compare flat plans at the same volume.
Estimate your bill
OpenRouter prices · no markup
Loading providers...
LLM cost on your key
$231.84/mo
~$1.1592 / review
Argus itself: $0
OpenRouter prices, live when the catalogue loads, at 322k tokens per review (95% input), one model for every stage. Deep Review costs more. Each review's usage block shows the real split.
Flat plans
at 200 reviews a month
Your key · Anthropic: Claude Sonnet 4.5
A month: $231.84
A review: $1.16
Argus's GitHub reviews: Yes · the estimate above
Claude Pro (opens in a new tab)
A month: $20
A review: $0.10
Argus's GitHub reviews: No · local review in Claude Code
Claude Max 5x (opens in a new tab)
A month: $100
A review: $0.50
Argus's GitHub reviews: No · local review in Claude Code
Claude Max 20x (opens in a new tab)
A month: $200
A review: $1.00
Argus's GitHub reviews: No · local review in Claude Code
ChatGPT Plus (opens in a new tab)
A month: $20
A review: $0.10
Argus's GitHub reviews: No · local review in Codex
ChatGPT Pro 5x (opens in a new tab)
A month: $100
A review: $0.50
Argus's GitHub reviews: No · local review in Codex
ChatGPT Pro 20x (opens in a new tab)
A month: $200
A review: $1.00
Argus's GitHub reviews: No · local review in Codex
OpenAI paused new sign-ups to this tier on 10 September 2026.
SuperGrok Heavy (opens in a new tab)
A month: $300
A review: $1.50
Argus's GitHub reviews: No · local review in Grok Build
xAI's pricing page lists this plan without a price; $300 is the figure pricing trackers report.
Z.ai GLM Coding Plan Max (opens in a new tab)
A month: $168
A review: $0.84
Argus's GitHub reviews: Yes · Zhipu key
Set the Zhipu key's base URL to https://api.z.ai/api/coding/paas/v4. Z.ai scopes the plan to coding tools, so check that its terms cover a review bot.
DeepSeek V4 Pro (opens in a new tab)
A month: $93.51
A review: $0.47
Argus's GitHub reviews: Yes · DeepSeek key
DeepSeek sells no plan. This is its V4 Pro API at peak rates on a cache miss; off-peak rates are half.
List prices, billed monthly, read in September 2026 from the page each plan links to unless its row says otherwise. A plan's per-review figure is its price divided by your volume, and holds only while the plan's usage caps cover that volume; vendors don't state the caps in tokens.
The estimate assumes 322k tokens per review, the median per-repo average in the review data behind Argus's size limits. Reviews are input-heavy: the diff and its context go in, and findings come out.
Each review Argus posts lists its tokens per pipeline stage, and the cost when your provider reports it or the model’s price is known, so you can check the estimate against your own reviews.
Claude, ChatGPT and SuperGrok plans can’t run Argus's GitHub reviews: none includes API access, and Anthropic doesn’t allow claude.ai login in third-party products. Each includes its vendor’s coding agent, which can review a branch on your machine with Argus's memory over MCP. Z.ai’s coding plan and DeepSeek’s API work as Argus provider keys.
Questions
What does it cost?
Argus itself is free and open source. You run it, so you pay your LLM and embeddings providers for tokens on your own keys, Clerk at its own rates for dashboard sign-in, and the Postgres and server it runs on.
Does it work with monorepos?
Yes. Argus triages each changed file on its own: lockfiles, vendored and generated files are skipped, small or config files get a light pass, the rest get a full review. By default a PR with 60 or more files or 1,500 or more changed lines is reviewed at reduced depth over its riskiest files, and one with 400 or more files or 20,000 or more lines is refused with a reason. Both limits are set for the whole org in Settings → Limits.
What data does Argus store?
Your Argus database keeps PR metadata, findings with a code snippet each, team memory with its embeddings, and each review run's saved state, which includes the diff, so an interrupted review can resume. The code graph stores symbol names and line ranges, not source. Live progress events are pruned after 7 days. All of it stays in your Postgres.
Can I use my own LLM API key?
Yes, it is the only setup. The dashboard takes keys for 11 providers: OpenRouter, OpenAI, Anthropic, Fireworks, Groq, Together, DeepSeek, Azure, AWS Bedrock, Zhipu and Vercel AI Gateway. Each pipeline stage can use its own model. Keys are stored AES-256-GCM encrypted.
What if Argus flags something incorrectly?
React thumbs-down or reply. A dismissal from someone with write access becomes team memory for that repo, and with an embeddings endpoint, a later finding that closely matches it is dropped or downgraded before it posts. Security findings and a short list of permanent checks can be downgraded that way but are never dropped.
Are any features held back?
No. There is no paid tier and nothing is gated. Every feature is a setting. Most are on by default; Deep Review (four specialist passes per file) is off by default because it costs more tokens.
Is there a hosted version?
No. Argus is self-hosted only right now. Self-hosting needs Postgres with pgvector 0.8.2 or later, your own GitHub App, a Clerk instance for dashboard sign-in, your LLM key and, for memory, an embeddings endpoint; the self-hosting guide walks through it.