PcapAnalyzer is a free self-hosted pcap analyzer with 27 detectors

PcapAnalyzer is a GitHub project that analyses network captures. You upload a .pcap or .pcapng file (gzip also works) and the tool reports what is failing, why, and what to check next. Its tagline promises to find the network fault "in seconds, not hours". It runs on your own machine, so captures never leave your network. It is free and open source under Apache 2.0, and its status is beta.
The core is 27 detectors that run on every capture, each reporting a severity and a next step. They are grouped by area. TCP: firewall blocks (SYN followed by RST), retransmissions, duplicate ACKs, loss and reordering, zero and full windows, resets after the handshake, and slow handshakes. Path: PMTUD black holes, low MSS, IP fragmentation, ICMP errors and routing loops. Name and address services: unanswered or slow DNS, SERVFAIL, NXDOMAIN bursts, DHCP NAK, DECLINE or no answer, duplicate IPs and ARP storms. Applications: HTTP 4xx and 5xx errors and slow responses, TLS alerts, and SSL 3.0, TLS 1.0 and TLS 1.1. Tunnels: IPsec (IKE and ESP), WireGuard, OpenVPN, GRE, L2TP and Teredo that are one-way, never answer a handshake or keep renegotiating. Security: port scans and sweeps, cleartext logins (FTP, POP3, IMAP, SMTP AUTH, HTTP Basic), Telnet, SMBv1, SNMP v1 and v2c, ARP spoofing and overlapping IP fragments. For cleartext logins, credentials are never stored or shown.
The README puts capture quality first. The tool warns when the capture itself limits the result: frames duplicated by a mirror port, traffic seen in one direction only, truncated packets, and connections that were already open when the capture started.
There are also Wireshark-style views: a packet list with real Wireshark display filters over the whole capture, colouring, CSV export, expert info, HTTP, DNS, DHCP and TLS tables, response times, and a flow diagram for each conversation. The interface comes in Spanish and English with light and dark themes, and the tool can produce a PDF report.
The AI investigation is optional. An AI examines the capture using tools (real filters, stream detail, statistics) and cites the packets behind every claim. You can then ask follow-up questions. It uses your own Anthropic API key; without one, a demo mode runs the full flow with text generated from the analysis, at no cost. The AI sees only what its tools return, which is packet metadata, plus packet content only if you switch that on for the session. It never sees the pcap file itself. While an investigation lasts, the capture is kept encrypted and is deleted after 60 minutes of inactivity or when you press "Delete capture now". The self-hosted version has no telemetry and no accounts.
Installing needs Docker with Docker Compose; on Windows the script runs from Git Bash or WSL. You clone the repository and run ./start.sh. The script creates a random access token, shown once and stored in .env, and starts the app. When the build finishes, you open http://localhost:8080, paste the token into the "API token" field at top right, and upload a capture or click "Try the sample capture". A hosted demo is available without installing anything, and a free download is offered on Gumroad as an alternative to cloning. A development setup needs Python 3.9+, Node 20+ and tshark, with the backend served by uvicorn and the frontend dev server on http://localhost:5173.
The author is candid that this is a beta. It has been tested with synthetic fixtures and public sample captures, and the most useful feedback is a capture where it misses a problem or reports one that is not real. Reports go to the GitHub issues page, with a request not to attach captures that contain real credentials or customer data. Detector thresholds, the JSON schema, limits and API errors are documented in README.es.md, in Spanish. Third-party tshark and Wireshark (GPL) run as a separate program and are not distributed with PcapAnalyzer.
Key facts
- PcapAnalyzer analyses .pcap and .pcapng captures (gzip too) with 27 detectors, each giving a severity and a next step.
- Detectors cover TCP, path, name and address services, applications, tunnels and security, such as firewall blocks, PMTUD black holes, DNS and DHCP failures, port scans and ARP spoofing.
- It runs locally under Apache 2.0 with no telemetry and no accounts; the app listens on 127.0.0.1 only. Status is beta.
- The optional AI investigation uses the user's own Anthropic API key, cites the packets behind its claims and never sees the pcap file; without a key a demo mode runs.
- Install is ./start.sh with Docker Compose, then http://localhost:8080 and a one-time access token.
Why it matters
Reading a capture by hand is slow, and PcapAnalyzer aims to shorten that: it runs 27 detectors on every upload and says what is failing, why, and what to check next. It also warns when the capture itself limits the result, for example mirror-port duplicates, one-direction traffic, truncated packets or connections already open at capture start. Everything stays on your own machine, which matters when a capture holds sensitive traffic.
Who it affects
Anyone with a .pcap or .pcapng file to make sense of: the detectors span TCP, DNS and DHCP, VPN tunnels and security patterns, and the Wireshark-style views let a reader drill into packets with real display filters. It suits people who want a self-hosted tool with the capture kept on their own network. The interface is available in Spanish and English.
How to use it
The easiest start is the hosted demo, which needs no install. For a local install you need Docker with Docker Compose (on Windows, Git Bash or WSL). Clone the repository, run ./start.sh, note the access token the script prints once, open http://localhost:8080, paste the token into the "API token" field and upload a capture or click "Try the sample capture". The project is free and open source under Apache 2.0, and a free download is also offered on Gumroad. A development setup needs Python 3.9+, Node 20+ and tshark. The AI investigation is optional: add your own Anthropic API key, or use the demo mode, which produces text from the analysis at no cost.
How solid is it
The project calls itself a beta. It has been tested with synthetic fixtures and public sample captures. No benchmark, accuracy figure, false-positive rate or performance measurement is given, and "in seconds, not hours" is a tagline, not a measurement. The source does not name the author or any organisation behind the project beyond the repository URL.
Risks and caveats
As a beta, it may miss problems or report ones that are not real; the author asks for exactly those cases as feedback. The app listens on 127.0.0.1 only and should not be exposed on a network without an authenticated reverse proxy, because the API can use your AI account. Captures can hold real credentials or customer data, so the README asks that any capture attached to an issue be anonymised first. During an AI investigation the capture is kept encrypted for up to 60 minutes of inactivity unless you delete it sooner. Detector thresholds and the JSON schema are documented only in the Spanish README.es.md. tshark and Wireshark are GPL and are not distributed with the project.
“Find the network fault in seconds, not hours.”
— PcapAnalyzer README