private-gpt: init at 0.5.0

nixos/doc/manual/release-notes/rl-2405.section.md

@@ -209,6 +209,8 @@ The pre-existing [services.ankisyncd](#opt-services.ankisyncd.enable) has been m
 
 - [isolate](https://github.com/ioi/isolate), a sandbox for securely executing untrusted programs. Available as [security.isolate](#opt-security.isolate.enable).
 
+- [private-gpt](https://github.com/zylon-ai/private-gpt), a service to interact with your documents using the power of LLMs, 100% privately, no data leaks. Available as [services.private-gpt](#opt-services.private-gpt.enable).
+
 ## Backward Incompatibilities {#sec-release-24.05-incompatibilities}
 
 <!-- To avoid merge conflicts, consider adding your item at an arbitrary place in the list instead. -->

nixos/modules/module-list.nix

@@ -776,6 +776,7 @@
   ./services/misc/polaris.nix
   ./services/misc/portunus.nix
   ./services/misc/preload.nix
+  ./services/misc/private-gpt.nix
   ./services/misc/prowlarr.nix
   ./services/misc/pufferpanel.nix
   ./services/misc/pykms.nix

nixos/modules/services/misc/private-gpt.nix

@@ -0,0 +1,121 @@
+{ config
+, lib
+, pkgs
+, ...
+}:
+let
+  inherit (lib) types;
+
+  format = pkgs.formats.yaml { };
+  cfg = config.services.private-gpt;
+in
+{
+  options = {
+    services.private-gpt = {
+      enable = lib.mkEnableOption "private-gpt for local large language models";
+      package = lib.mkPackageOption pkgs "private-gpt" { };
+
+      stateDir = lib.mkOption {
+        type = types.path;
+        default = "/var/lib/private-gpt";
+        description = "State directory of private-gpt.";
+      };
+
+      settings = lib.mkOption {
+        type = format.type;
+        default = {
+          llm = {
+            mode = "ollama";
+            tokenizer = "";
+          };
+          embedding = {
+            mode = "ollama";
+          };
+          ollama = {
+            llm_model = "llama3";
+            embedding_model = "nomic-embed-text";
+            api_base = "http://localhost:11434";
+            embedding_api_base = "http://localhost:11434";
+            keep_alive = "5m";
+            tfs_z = 1;
+            top_k = 40;
+            top_p = 0.9;
+            repeat_last_n = 64;
+            repeat_penalty = 1.2;
+            request_timeout = 120;
+          };
+          vectorstore = {
+            database = "qdrant";
+          };
+          qdrant = {
+            path = "/var/lib/private-gpt/vectorstore/qdrant";
+          };
+          data = {
+            local_data_folder = "/var/lib/private-gpt";
+          };
+          openai = { };
+          azopenai = { };
+        };
+        description = ''
+          settings-local.yaml for private-gpt
+        '';
+      };
+    };
+  };
+
+  config = lib.mkIf cfg.enable {
+    systemd.services.private-gpt = {
+      description = "Interact with your documents using the power of GPT, 100% privately, no data leaks";
+      wantedBy = [ "multi-user.target" ];
+      after = [ "network.target" ];
+
+      preStart =
+        let
+          config = format.generate "settings-local.yaml" (cfg.settings // { server.env_name = "local"; });
+        in
+        ''
+          mkdir -p ${cfg.stateDir}/{settings,huggingface,matplotlib,tiktoken_cache}
+          cp ${cfg.package.cl100k_base.tiktoken} ${cfg.stateDir}/tiktoken_cache/9b5ad71b2ce5302211f9c61530b329a4922fc6a4
+          cp ${pkgs.python3Packages.private-gpt}/${pkgs.python3.sitePackages}/private_gpt/settings.yaml ${cfg.stateDir}/settings/settings.yaml
+          cp "${config}" "${cfg.stateDir}/settings/settings-local.yaml"
+          chmod 600 "${cfg.stateDir}/settings/settings-local.yaml"
+        '';
+
+      environment = {
+        PGPT_PROFILES = "local";
+        PGPT_SETTINGS_FOLDER = "${cfg.stateDir}/settings";
+        HF_HOME = "${cfg.stateDir}/huggingface";
+        TRANSFORMERS_OFFLINE = "1";
+        HF_DATASETS_OFFLINE = "1";
+        MPLCONFIGDIR = "${cfg.stateDir}/matplotlib";
+      };
+
+      serviceConfig = {
+        ExecStart = lib.getExe cfg.package;
+        WorkingDirectory = cfg.stateDir;
+        StateDirectory = "private-gpt";
+        RuntimeDirectory = "private-gpt";
+        RuntimeDirectoryMode = "0755";
+        PrivateTmp = true;
+        DynamicUser = true;
+        DevicePolicy = "closed";
+        LockPersonality = true;
+        MemoryDenyWriteExecute = true;
+        PrivateUsers = true;
+        ProtectHome = true;
+        ProtectHostname = true;
+        ProtectKernelLogs = true;
+        ProtectKernelModules = true;
+        ProtectKernelTunables = true;
+        ProtectControlGroups = true;
+        ProcSubset = "pid";
+        RestrictNamespaces = true;
+        RestrictRealtime = true;
+        SystemCallArchitectures = "native";
+        UMask = "0077";
+      };
+    };
+  };
+
+  meta.maintainers = with lib.maintainers; [ drupol ];
+}

nixos/tests/all-tests.nix

@@ -757,6 +757,7 @@ in {
   pretix = runTest ./web-apps/pretix.nix;
   printing-socket = handleTest ./printing.nix { socket = true; };
   printing-service = handleTest ./printing.nix { socket = false; };
+  private-gpt = handleTest ./private-gpt.nix {};
   privoxy = handleTest ./privoxy.nix {};
   prometheus = handleTest ./prometheus.nix {};
   prometheus-exporters = handleTest ./prometheus-exporters.nix {};

nixos/tests/private-gpt.nix

@@ -0,0 +1,27 @@
+import ./make-test-python.nix ({ pkgs, lib, ... }:
+let
+  mainPort = "8001";
+in
+{
+  name = "private-gpt";
+  meta = with lib.maintainers; {
+    maintainers = [ drupol ];
+  };
+
+  nodes = {
+    machine = { ... }: {
+      services.private-gpt = {
+        enable = true;
+      };
+    };
+  };
+
+  testScript = ''
+    machine.start()
+
+    machine.wait_for_unit("private-gpt.service")
+    machine.wait_for_open_port(${mainPort})
+
+    machine.succeed("curl http://127.0.0.1:${mainPort}")
+  '';
+})

pkgs/by-name/pr/private-gpt/package.nix

@@ -0,0 +1,17 @@
+{ python3Packages
+, makeBinaryWrapper
+}:
+
+python3Packages.toPythonApplication (python3Packages.private-gpt.overrideAttrs (oldAttrs: {
+  nativeBuildInputs = oldAttrs.nativeBuildInputs ++ [ makeBinaryWrapper ];
+
+  passthru.cl100k_base = {
+    inherit (python3Packages.private-gpt.cl100k_base) tiktoken;
+  };
+
+  postInstall = ''
+    makeWrapper ${python3Packages.python.interpreter} $out/bin/private-gpt \
+      --prefix PYTHONPATH : "$PYTHONPATH" \
+      --add-flags "-m private_gpt"
+  '';
+}))

pkgs/development/python-modules/private-gpt/default.nix

@@ -0,0 +1,119 @@
+{ lib
+, buildPythonPackage
+, python
+, fetchFromGitHub
+, poetry-core
+, fastapi
+, injector
+, llama-index-core
+, llama-index-readers-file
+, huggingface-hub
+, python-multipart
+, pyyaml
+, transformers
+, uvicorn
+, watchdog
+, gradio
+, fetchurl
+, fetchpatch
+}:
+
+buildPythonPackage rec {
+  pname = "private-gpt";
+  version = "0.5.0";
+  pyproject = true;
+
+  src = fetchFromGitHub {
+    owner = "zylon-ai";
+    repo = "private-gpt";
+    rev = "v${version}";
+    hash = "sha256-bjydzJhOJjmbflcJbuMyNsmby7HtNPFW3MY2Tw12cHw=";
+  };
+
+  patches = [
+    # Fix a vulnerability, to be removed in the next bump version
+    # See https://github.com/zylon-ai/private-gpt/pull/1890
+    (fetchpatch {
+      url = "https://github.com/zylon-ai/private-gpt/commit/86368c61760c9cee5d977131d23ad2a3e063cbe9.patch";
+      hash = "sha256-4ysRUuNaHW4bmNzg4fn++89b430LP6AzYDoX2HplVH0=";
+    })
+  ];
+
+  build-system = [
+    poetry-core
+  ];
+
+  dependencies = [
+    fastapi
+    injector
+    llama-index-core
+    llama-index-readers-file
+    python-multipart
+    pyyaml
+    transformers
+    uvicorn
+    watchdog
+  ] ++ lib.flatten (builtins.attrValues passthru.optional-dependencies);
+
+  # This is needed for running the tests and the service in offline mode,
+  # See related issue at https://github.com/zylon-ai/private-gpt/issues/1870
+  passthru.cl100k_base.tiktoken = fetchurl {
+    url = "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken";
+    hash = "sha256-Ijkht27pm96ZW3/3OFE+7xAPtR0YyTWXoRO8/+hlsqc=";
+  };
+
+  passthru.optional-dependencies = with python.pkgs; {
+    embeddings-huggingface = [
+      huggingface-hub
+      llama-index-embeddings-huggingface
+    ];
+    embeddings-ollama = [
+      llama-index-embeddings-ollama
+    ];
+    embeddings-openai = [
+      llama-index-embeddings-openai
+    ];
+    embeddings-sagemaker = [
+      boto3
+    ];
+    llms-ollama = [
+      llama-index-llms-ollama
+    ];
+    llms-openai = [
+      llama-index-llms-openai
+    ];
+    llms-openai-like = [
+      llama-index-llms-openai-like
+    ];
+    llms-sagemaker = [
+      boto3
+    ];
+    ui = [
+      gradio
+    ];
+    vector-stores-chroma = [
+      llama-index-vector-stores-chroma
+    ];
+    vector-stores-postgres = [
+      llama-index-vector-stores-postgres
+    ];
+    vector-stores-qdrant = [
+      llama-index-vector-stores-qdrant
+    ];
+  };
+
+  postInstall = ''
+    cp settings*.yaml $out/${python.sitePackages}/private_gpt/
+  '';
+
+  pythonImportsCheck = [ "private_gpt" ];
+
+  meta = {
+    changelog = "https://github.com/zylon-ai/private-gpt/blob/${src.rev}/CHANGELOG.md";
+    description = "Interact with your documents using the power of GPT, 100% privately, no data leaks";
+    homepage = "https://github.com/zylon-ai/private-gpt";
+    license = lib.licenses.asl20;
+    mainProgram = "private-gpt";
+    maintainers = with lib.maintainers; [ drupol ];
+  };
+}

pkgs/top-level/python-packages.nix

@@ -9891,6 +9891,8 @@ self: super: with self; {
 
   ppdeep = callPackage ../development/python-modules/ppdeep { };
 
+  private-gpt = callPackage ../development/python-modules/private-gpt { };
+
   prodict = callPackage ../development/python-modules/prodict { };
 
   prometheus-pandas = callPackage ../development/python-modules/prometheus-pandas { };