config

package
v0.0.0-...-0c5745b Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Apr 19, 2026 License: Apache-2.0 Imports: 20 Imported by: 0

Documentation

Overview

Package config provides configuration loading, merging, and persistence for the markedup CLI. Configuration cascades from global (~/.markedup.yaml) to local ({kbDir}/.markedup.yaml) to environment variables.

Package config provides configuration management for markedup, including auto-detection of local model serving endpoints.

Package config provides configuration management for markedup, including secure API key storage via OS-native keychains.

Package config provides configuration helpers for the markedup CLI, including model download utilities for Ollama and HuggingFace.

Index

Constants

View Source 
const ProbeTimeout = 10 * time.Second

ProbeTimeout is the per-request deadline for Test Connection probes. Kept generous because cold-loaded local models can take several seconds to respond to the first request.

Variables

This section is empty.

Functions

func DeleteKey 

func DeleteKey(name string) error

DeleteKey removes a named key from the OS keychain. Returns nil if the key did not exist.

func Exists 

func Exists(kbDir string) bool

Exists returns true if at least one config file (global or local) exists.

func GetKey 

func GetKey(name string) (string, error)

GetKey retrieves a named key from the OS keychain. Returns ("", nil) if the key is not found.

func GlobalPath 

func GlobalPath() string

GlobalPath returns the path to the global config file (~/.markedup.yaml).

func HFAvailable 

func HFAvailable() bool

HFAvailable reports whether the hf CLI binary is in PATH.

func HFPull 

func HFPull(ctx context.Context, model string, w io.Writer) error

HFPull runs "hf download <model>" and streams output to w. The process is terminated if ctx is cancelled.

func HydrateKeys 

func HydrateKeys(cfg *Config) error

HydrateKeys runs the one-time-per-process legacy-key migration and then fills any empty APIKey fields in cfg from the OS keyring, looked up by endpoint. It is safe to call multiple times in one process: migration is gated by sync.Once so it runs at most once, while hydration runs every call (different cfgs may have different endpoints).

Subcommands that make outbound API calls (enrich, serve, embed, tui) or that store credentials (setup) must call this after config.Load(). Local- only subcommands (check, init, search, explore, show, export) must not.

func IsNuExtractModel 

func IsNuExtractModel(id string) bool

IsNuExtractModel reports whether a model id names a NuExtract-2.0 model. Matches the canonical form `nuextract-2.0-*` (anchored on the trailing dash) with any `publisher/` prefix stripped. Case-insensitive. Examples matched: "nuextract-2.0-8b", "numind/NuExtract-2.0-2B", "someorg/nuextract-2.0-4b-gguf". Examples rejected: "nuextract-2.0", "nuextract-2.0foo", "nuextract-1.5-*", "nuextract2.0-8b".

func KeyNameForEndpoint 

func KeyNameForEndpoint(endpoint string) string

KeyNameForEndpoint returns the keyring entry name used to store the API key for the given endpoint. Format: "apikey-<16 hex chars of sha256(normalized)>". Returns "" if the endpoint is empty after normalization.

func KeyringAvailable 

func KeyringAvailable() bool

KeyringAvailable returns true if the OS keyring can be opened successfully. Returns false in headless/CI environments where no keychain backend is available.

func LocalPath 

func LocalPath(kbDir string) string

LocalPath returns the path to the local (per-KB) config file.

func NormalizeEndpoint 

func NormalizeEndpoint(endpoint string) string

NormalizeEndpoint canonicalizes an endpoint URL for keyring keying. Lowercases the host, strips default ports (80/443), and removes trailing slashes from the path so trivially equivalent URLs hash identically. If the input cannot be parsed as a URL, the trimmed lowercase string is returned as a stable fallback.

func OllamaAvailable 

func OllamaAvailable() bool

OllamaAvailable reports whether the ollama CLI binary is in PATH.

func OllamaPull 

func OllamaPull(ctx context.Context, model string, w io.Writer) error

OllamaPull runs "ollama pull <model>" and streams output to w. The process is terminated if ctx is cancelled.

func Save 

func Save(cfg *Config, path string) error

Save writes cfg to path as YAML. Fields tagged yaml:"-" (APIKey) are automatically omitted by the YAML encoder.

func StoreKey 

func StoreKey(name, value string) error

StoreKey saves a named key to the OS keychain under the "markedup" service. The name should be produced by KeyNameForEndpoint so multiple services sharing one endpoint reuse the same keychain entry.

Types

type Config 

type Config struct {
	Embed     ServiceConfig   `yaml:"embed"`
	Rerank    RerankConfig    `yaml:"rerank"`
	LLM       ServiceConfig   `yaml:"llm"`
	Triplex   ServiceConfig   `yaml:"triplex,omitempty"`
	NuExtract NuExtractConfig `yaml:"nuextract,omitempty"`
	Enrich    EnrichConfig    `yaml:"enrich,omitempty"`
	// Format selects the Tier 2 extractor: "triplex", "nuextract", "generic", or "" (auto).
	Format string `yaml:"format,omitempty"`
}

Config holds all service configurations for markedup.

func Load 

func Load(kbDir string) (*Config, error)

Load reads the global and local config files, merges them (local overrides global), then applies environment variable overrides. A missing file is not an error — the cascade simply skips it.

type Endpoint 

type Endpoint struct {
	Name    string // "Ollama", "LM Studio", "TEI", "vLLM", "llama.cpp"
	URL     string // e.g. "http://localhost:11434"
	Type    string // "embed", "llm", "rerank", "multi"
	Healthy bool
	Models  []string // populated when the service exposes a models API
	// Formats lists Tier 2 extractors this endpoint can serve (e.g. "nuextract"
	// when a NuExtract-2.0 model id is present in Models).
	Formats []string
}

Endpoint describes a discovered local model-serving endpoint.

func DetectLocal 

func DetectLocal(ctx context.Context) []Endpoint

DetectLocal probes well-known localhost ports concurrently and returns all discovered endpoints sorted by name. It never returns an error; unreachable services are silently skipped.

type EnrichConfig 

type EnrichConfig struct {
	Fallback FallbackConfig `yaml:"fallback,omitempty"`
}

EnrichConfig holds enrich-pipeline-wide settings. Currently a single nested Fallback block; left as a struct (not inlined) so future enrich settings — Parallel, Verbose, FailOnError — can land without churning the top-level config shape.

type FallbackConfig 

type FallbackConfig struct {
	// Enabled is a tri-state pointer: nil = "use the local/cloud default",
	// non-nil = explicit user choice. Local endpoints default-on; cloud
	// endpoints default-off (cost guard).
	Enabled *bool `yaml:"enabled,omitempty"`
	// Endpoint overrides cfg.LLM.Endpoint for fallback only. Empty = inherit.
	Endpoint string `yaml:"endpoint,omitempty"`
	// Model overrides cfg.LLM.Model for fallback only. Empty = inherit.
	Model string `yaml:"model,omitempty"`
	// APIKey overrides cfg.LLM.APIKey for fallback only.
	APIKey string `yaml:"-"` // NEVER written to YAML
	// MaxFiles caps the number of files dispatched per enrich run. Nil = use
	// the local/cloud default (50 cloud, unbounded local). Explicit 0 = unbounded.
	MaxFiles *int `yaml:"max_files,omitempty"`
	// Parallel enables concurrent fallback LLM calls (#141). Default false —
	// sequential execution remains the safe default. Opt-in for users with
	// capable hardware or cloud endpoints with high RPS.
	Parallel bool `yaml:"parallel,omitempty"`
	// ParallelWorkers caps the worker pool when Parallel=true. <=0 falls back
	// to runtime.NumCPU() at dispatch time.
	ParallelWorkers int `yaml:"parallel_workers,omitempty"`
}

FallbackConfig configures the LLM fallback dispatcher (issue #140).

Defaults are resolved at call-time, not parse-time: an empty Endpoint inherits cfg.LLM.Endpoint (the same env-var-backed reasoning endpoint the markedup_reason MCP tool uses), and Enabled / MaxFiles defaults depend on whether the resolved endpoint is local or cloud. See enrich.IsLocalEndpoint and the resolution path in internal/cli/enrich.go.

type MigrationResult 

type MigrationResult struct {
	Migrated  []string // legacy names successfully copied + removed
	Conflicts []string // legacy names left in place because the endpoint-keyed entry already held a different value
	Skipped   []string // legacy names with no endpoint configured (deferred)
}

MigrationResult summarizes what MigrateLegacyKeys did, primarily for tests and logging. Counts are zero on a no-op (no legacy entries present).

func MigrateLegacyKeys 

func MigrateLegacyKeys(cfg *Config) MigrationResult

MigrateLegacyKeys consolidates legacy per-service keyring entries ("embed-api-key", "llm-api-key", etc.) into the new endpoint-keyed format.

Behavior:

  • For each legacy entry that exists, look up the configured endpoint for that service. If the endpoint is unknown, leave the legacy entry alone (a later run, with a populated config, can migrate it).
  • If no endpoint-keyed entry exists yet, write the legacy value there and delete the legacy entry.
  • If an endpoint-keyed entry exists with the SAME value, the legacy entry is a duplicate; delete it.
  • If an endpoint-keyed entry exists with a DIFFERENT value, this is a conflict. Preserve BOTH entries (do not delete the legacy one) and emit a warning so the user can reconcile via the setup wizard. Idempotent on re-run: same situation -> same warning, no destructive action.

Idempotent: repeated invocations after a successful migration are no-ops because the legacy entries no longer exist; conflict cases stay stable too.

type NuExtractConfig 

type NuExtractConfig struct {
	ServiceConfig `yaml:",inline"`
	Mode          string   `yaml:"mode,omitempty"`
	Transport     string   `yaml:"transport,omitempty"`
	Predicates    []string `yaml:"predicates,omitempty"`
	EntityTypes   []string `yaml:"entity_types,omitempty"`
}

NuExtractConfig configures the NuExtract-2.0 template-based extractor. Mode controls the two-pass runner: "parallel" (default) fires entities and relations calls simultaneously; "single" sends one combined template. Transport selects request shape: "native" uses chat_template_kwargs for vLLM/HF; "manual" renders the prompt client-side for GGUF runtimes. Empty Transport auto-detects from the endpoint URL.

type ProbeResult 

type ProbeResult struct {
	OK     bool
	Detail string
}

ProbeResult is returned by every Probe* function. OK reflects success (HTTP 200 + a structurally valid response). Detail provides a short human-readable summary suitable for inline display in the wizard (either an error reason or a confirmation snippet).

func ProbeChat 

func ProbeChat(ctx context.Context, endpoint, model, apiKey string) ProbeResult

ProbeChat sends a tiny chat-completion to <endpoint>/v1/chat/completions and verifies the model returns a non-empty response. Used for LLM, Triplex, and NuExtract endpoints (all OpenAI-compatible chat).

func ProbeEmbed 

func ProbeEmbed(ctx context.Context, endpoint, model, apiKey string) ProbeResult

ProbeEmbed sends a tiny embedding request to <endpoint>/v1/embeddings and verifies a non-empty vector is returned.

func ProbeRerank 

func ProbeRerank(ctx context.Context, endpoint, model, apiKey string) ProbeResult

ProbeRerank sends a tiny rerank request to <endpoint>/v1/rerank and verifies a 200 response. The shape of rerank responses varies between providers, so we accept any 200 with a parseable JSON body.

func ProbeService 

func ProbeService(ctx context.Context, service, endpoint, model, apiKey string) ProbeResult

ProbeService dispatches to the appropriate probe function based on service name. Recognized names: "embed", "llm", "triplex", "nuextract", "rerank". Unknown service names return a failed result.

type RerankConfig 

type RerankConfig struct {
	ServiceConfig `yaml:",inline"`
	Format        string `yaml:"format,omitempty"`
}

RerankConfig extends ServiceConfig with a reranker-specific format field.

type ServiceConfig 

type ServiceConfig struct {
	Endpoint string `yaml:"endpoint"`
	Model    string `yaml:"model"`
	APIKey   string `yaml:"-"` // NEVER written to YAML
}

ServiceConfig holds endpoint, model, and authentication for a single service.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.