mirror of
https://github.com/BaaRai/ARTEX.git
synced 2026-11-13 11:40:53 +08:00
feat(llm): optional custom User-Agent for LLM requests
Add an optional per-profile User-Agent that overrides the header on every outbound LLM request. When empty (the default), the existing behaviour is unchanged and the default User-Agent is left untouched. The value is trimmed and validated: control characters / newlines (header injection) and values longer than 256 bytes are rejected. Wired through the profile config struct, the Postgres llm_profiles schema, the profile API/DTO, the connection-test path, and the LLM profile form, following the existing session-header field. Addresses part of #154; the fixed-full-URL mode is not included.
This commit is contained in:
BaaRai
@@ -2,6 +2,12 @@
|
||||
|
||||
All notable changes to this project are documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
|
||||
|
||||
## [Unreleased]
|
||||
|
||||
### Added
|
||||
|
||||
- **Optional custom User-Agent for LLM requests**: an LLM profile can set a `User-Agent` value that overrides the header on every outbound LLM request; left empty, the default User-Agent is unchanged. The value is trimmed and rejected if it contains control characters or exceeds 256 bytes. Part of #154 (the fixed-full-URL mode is not included).
|
||||
|
||||
## [0.5.1] - 2026-11-05
|
||||
|
||||
Login rate limiting per client IP with trusted-proxy support, a 12-character minimum for new passwords, a loopback-only default for the binary, and the JWT signing key in its own Docker volume.
|
||||
|
||||
+39
-3
@@ -86,6 +86,10 @@ type Config struct {
|
||||
// context via transcript.WithSessionID and read by the RoundTripper, so even a
|
||||
// shared provider emits different header values per session.
|
||||
SessionHeaderKey string
|
||||
// UserAgent, when non-empty, overrides the User-Agent header on every outbound
|
||||
// LLM request built from this config. Empty = don't set it, leaving the Go/Norma
|
||||
// default UA untouched (the historical behavior). Validated by ValidateUserAgent.
|
||||
UserAgent string
|
||||
// Retry holds the retry parameters resolved for this config (profile override →
|
||||
// global policy → built-in default, resolved server-side). See RetryConfig for
|
||||
// the per-layer semantics; zero values = fully inherit the built-in defaults.
|
||||
@@ -341,7 +345,7 @@ func (c Config) Provider() string {
|
||||
// limiter lives on the single provider instance — so planner + all workers +
|
||||
// main agent (which share this provider) are bounded by one shared rate limit.
|
||||
func (c Config) NewProvider() (llm.Provider, error) {
|
||||
client, err := quotaAwareHTTPClient(c.Proxy, c.SessionHeaderKey)
|
||||
client, err := quotaAwareHTTPClient(c.Proxy, c.SessionHeaderKey, c.UserAgent)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
@@ -440,6 +444,9 @@ type quotaAwareTransport struct {
|
||||
// carries; its value is the session id read from the request context. Empty
|
||||
// disables it. See Config.SessionHeaderKey.
|
||||
sessionHeaderKey string
|
||||
// userAgent, when non-empty, is set as the User-Agent header on every request.
|
||||
// Empty leaves the default UA untouched. See Config.UserAgent.
|
||||
userAgent string
|
||||
}
|
||||
|
||||
func (t quotaAwareTransport) RoundTrip(req *http.Request) (*http.Response, error) {
|
||||
@@ -452,6 +459,11 @@ func (t quotaAwareTransport) RoundTrip(req *http.Request) (*http.Response, error
|
||||
req.Header.Set(t.sessionHeaderKey, sid)
|
||||
}
|
||||
}
|
||||
// Custom User-Agent: override only when configured, so an empty value keeps the
|
||||
// underlying transport's default UA exactly as before.
|
||||
if t.userAgent != "" {
|
||||
req.Header.Set("User-Agent", t.userAgent)
|
||||
}
|
||||
// When LLM recording is on, the Recorder puts a Capture on the context so the
|
||||
// raw wire bodies can be persisted. This is the only layer that still sees
|
||||
// them: norma builds the request body internally and decodes the SSE response
|
||||
@@ -504,7 +516,7 @@ func requestBodySnapshot(req *http.Request) string {
|
||||
return string(b)
|
||||
}
|
||||
|
||||
func quotaAwareHTTPClient(proxy, sessionHeaderKey string) (*http.Client, error) {
|
||||
func quotaAwareHTTPClient(proxy, sessionHeaderKey, userAgent string) (*http.Client, error) {
|
||||
transport := http.DefaultTransport.(*http.Transport).Clone()
|
||||
proxy = strings.TrimSpace(proxy)
|
||||
if proxy == "" {
|
||||
@@ -523,7 +535,31 @@ func quotaAwareHTTPClient(proxy, sessionHeaderKey string) (*http.Client, error)
|
||||
}
|
||||
transport.Proxy = http.ProxyURL(proxyURL)
|
||||
}
|
||||
return &http.Client{Transport: quotaAwareTransport{base: transport, sessionHeaderKey: strings.TrimSpace(sessionHeaderKey)}}, nil
|
||||
return &http.Client{Transport: quotaAwareTransport{base: transport, sessionHeaderKey: strings.TrimSpace(sessionHeaderKey), userAgent: strings.TrimSpace(userAgent)}}, nil
|
||||
}
|
||||
|
||||
// MaxUserAgentLen bounds a custom User-Agent value. Generous for real UA strings
|
||||
// while keeping a pathological value out of every outbound request.
|
||||
const MaxUserAgentLen = 256
|
||||
|
||||
// ValidateUserAgent trims a custom User-Agent and returns the cleaned value. An
|
||||
// empty/blank input is valid and means "no override" (keep the default UA). A
|
||||
// non-empty value is rejected when it exceeds MaxUserAgentLen or contains any
|
||||
// control character (including CR/LF, which would allow header injection).
|
||||
func ValidateUserAgent(raw string) (string, error) {
|
||||
ua := strings.TrimSpace(raw)
|
||||
if ua == "" {
|
||||
return "", nil
|
||||
}
|
||||
if len(ua) > MaxUserAgentLen {
|
||||
return "", fmt.Errorf("user agent too long: %d bytes (max %d)", len(ua), MaxUserAgentLen)
|
||||
}
|
||||
for _, r := range ua {
|
||||
if r < 0x20 || r == 0x7f {
|
||||
return "", fmt.Errorf("user agent contains a control character")
|
||||
}
|
||||
}
|
||||
return ua, nil
|
||||
}
|
||||
|
||||
// logTestConnection prints the raw HTTP status code(s) and response body of a
|
||||
|
||||
@@ -21,7 +21,7 @@ func TestRoundTripCapturesRawBodies(t *testing.T) {
|
||||
}))
|
||||
defer srv.Close()
|
||||
|
||||
client, err := quotaAwareHTTPClient("", "")
|
||||
client, err := quotaAwareHTTPClient("", "", "")
|
||||
if err != nil {
|
||||
t.Fatalf("client: %v", err)
|
||||
}
|
||||
@@ -65,7 +65,7 @@ func TestRoundTripCaptures429BodyAlongsideQuotaRewrite(t *testing.T) {
|
||||
}))
|
||||
defer srv.Close()
|
||||
|
||||
client, err := quotaAwareHTTPClient("", "")
|
||||
client, err := quotaAwareHTTPClient("", "", "")
|
||||
if err != nil {
|
||||
t.Fatalf("client: %v", err)
|
||||
}
|
||||
@@ -106,7 +106,7 @@ func TestRoundTripWithoutCaptureIsUnchanged(t *testing.T) {
|
||||
}))
|
||||
defer srv.Close()
|
||||
|
||||
client, err := quotaAwareHTTPClient("", "")
|
||||
client, err := quotaAwareHTTPClient("", "", "")
|
||||
if err != nil {
|
||||
t.Fatalf("client: %v", err)
|
||||
}
|
||||
|
||||
@@ -0,0 +1,68 @@
|
||||
package agent
|
||||
|
||||
import (
|
||||
"net/http"
|
||||
"net/http/httptest"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
// serveUA spins up an httptest server that records the User-Agent of the request
|
||||
// it receives, drives one request through a quotaAwareHTTPClient built with the
|
||||
// given userAgent, and returns what actually reached the wire.
|
||||
func serveUA(t *testing.T, userAgent string) string {
|
||||
t.Helper()
|
||||
got := make(chan string, 1)
|
||||
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
got <- r.Header.Get("User-Agent")
|
||||
w.WriteHeader(http.StatusOK)
|
||||
}))
|
||||
defer srv.Close()
|
||||
|
||||
client, err := quotaAwareHTTPClient("", "", userAgent)
|
||||
if err != nil {
|
||||
t.Fatalf("quotaAwareHTTPClient: %v", err)
|
||||
}
|
||||
resp, err := client.Get(srv.URL)
|
||||
if err != nil {
|
||||
t.Fatalf("GET: %v", err)
|
||||
}
|
||||
resp.Body.Close()
|
||||
return <-got
|
||||
}
|
||||
|
||||
func TestUserAgentHeaderSetWhenConfigured(t *testing.T) {
|
||||
const want = "artex-test-agent/1.2.3"
|
||||
if got := serveUA(t, want); got != want {
|
||||
t.Fatalf("User-Agent = %q, want %q", got, want)
|
||||
}
|
||||
}
|
||||
|
||||
func TestUserAgentHeaderDefaultWhenEmpty(t *testing.T) {
|
||||
// Empty = no override: the Go client's default UA must reach the server unchanged.
|
||||
got := serveUA(t, "")
|
||||
if !strings.HasPrefix(got, "Go-http-client/") {
|
||||
t.Fatalf("User-Agent = %q, want the default Go-http-client UA when unset", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestValidateUserAgentRejectsNewline(t *testing.T) {
|
||||
if _, err := ValidateUserAgent("good/1.0\r\nX-Injected: 1"); err == nil {
|
||||
t.Fatalf("expected a newline-containing user agent to be rejected")
|
||||
}
|
||||
}
|
||||
|
||||
func TestValidateUserAgentRejectsTooLong(t *testing.T) {
|
||||
if _, err := ValidateUserAgent(strings.Repeat("a", MaxUserAgentLen+1)); err == nil {
|
||||
t.Fatalf("expected a user agent longer than %d to be rejected", MaxUserAgentLen)
|
||||
}
|
||||
}
|
||||
|
||||
func TestValidateUserAgentTrimsAndAllowsEmpty(t *testing.T) {
|
||||
if ua, err := ValidateUserAgent(" "); err != nil || ua != "" {
|
||||
t.Fatalf("blank user agent: got (%q, %v), want (\"\", nil)", ua, err)
|
||||
}
|
||||
if ua, err := ValidateUserAgent(" my-app/1.0 "); err != nil || ua != "my-app/1.0" {
|
||||
t.Fatalf("trim: got (%q, %v), want (\"my-app/1.0\", nil)", ua, err)
|
||||
}
|
||||
}
|
||||
+15
-10
@@ -64,6 +64,10 @@ type LLMProfile struct {
|
||||
// or sticky routing off a session-id header. "" = not sent. Maps to
|
||||
// agent.Config.SessionHeaderKey.
|
||||
SessionHeaderKey string `json:"session_header_key"`
|
||||
// UserAgent, when non-empty, overrides the User-Agent header on every outbound
|
||||
// request built from this profile. "" = don't override (keep the default UA);
|
||||
// old profiles behave exactly as before. Maps to agent.Config.UserAgent.
|
||||
UserAgent string `json:"user_agent"`
|
||||
// Cost ledger: EUR per 1M tokens (0 = unset; cost is computed only when a
|
||||
// price is set). Set through the same profile API/SaveProfile path.
|
||||
PriceInputPerM float64 `json:"price_input_per_m,omitempty"`
|
||||
@@ -89,8 +93,8 @@ type RetryOverride struct {
|
||||
// profileCols is the read column list (hint variant, no api key) shared by the
|
||||
// list query; profileColsKey is the same with api_key for the single-row loads.
|
||||
const profileRetryCols = `COALESCE(retry_connect_attempts,0),COALESCE(retry_connect_interval_ms,0),COALESCE(retry_empty_attempts,0),COALESCE(retry_empty_interval_ms,0),COALESCE(retry_stream_attempts,0),COALESCE(retry_stream_interval_ms,0)`
|
||||
const profileCols = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key_hint,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,''),COALESCE(session_header_key,''),` + profileRetryCols + `,COALESCE(price_input_per_m,0),COALESCE(price_output_per_m,0),COALESCE(price_cache_read_per_m,0),COALESCE(price_cache_write_per_m,0)`
|
||||
const profileColsKey = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,''),COALESCE(session_header_key,''),` + profileRetryCols + `,COALESCE(price_input_per_m,0),COALESCE(price_output_per_m,0),COALESCE(price_cache_read_per_m,0),COALESCE(price_cache_write_per_m,0)`
|
||||
const profileCols = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key_hint,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,''),COALESCE(session_header_key,''),` + profileRetryCols + `,COALESCE(price_input_per_m,0),COALESCE(price_output_per_m,0),COALESCE(price_cache_read_per_m,0),COALESCE(price_cache_write_per_m,0),COALESCE(user_agent,'')`
|
||||
const profileColsKey = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,''),COALESCE(session_header_key,''),` + profileRetryCols + `,COALESCE(price_input_per_m,0),COALESCE(price_output_per_m,0),COALESCE(price_cache_read_per_m,0),COALESCE(price_cache_write_per_m,0),COALESCE(user_agent,'')`
|
||||
|
||||
// scanProfile reads one row in the profileCols / profileColsKey column order. The
|
||||
// 7th column lands in APIKeyHint or APIKey depending on which list the caller used.
|
||||
@@ -101,7 +105,8 @@ func scanProfile(sc interface{ Scan(...any) error }, into *string, p *LLMProfile
|
||||
&p.Retry.Connect.Attempts, &p.Retry.Connect.IntervalMS,
|
||||
&p.Retry.Empty.Attempts, &p.Retry.Empty.IntervalMS,
|
||||
&p.Retry.Stream.Attempts, &p.Retry.Stream.IntervalMS,
|
||||
&p.PriceInputPerM, &p.PriceOutputPerM, &p.PriceCacheReadPerM, &p.PriceCacheWritePerM)
|
||||
&p.PriceInputPerM, &p.PriceOutputPerM, &p.PriceCacheReadPerM, &p.PriceCacheWritePerM,
|
||||
&p.UserAgent)
|
||||
}
|
||||
|
||||
func (d *DB) ListProfiles() ([]*LLMProfile, error) {
|
||||
@@ -180,27 +185,27 @@ func (d *DB) SaveProfile(p *LLMProfile) (int64, error) {
|
||||
r := p.Retry.Clamped()
|
||||
if p.ID == 0 {
|
||||
var id int64
|
||||
err := d.QueryRow(`INSERT INTO llm_profiles(name,format,base_url,proxy,model,api_key,api_key_hint,rate_per_second,rate_per_minute,context_window_k,reasoning_effort,priority,pool_exclude,thinking_type,streaming,max_tokens,max_tokens_field,session_header_key,retry_connect_attempts,retry_connect_interval_ms,retry_empty_attempts,retry_empty_interval_ms,retry_stream_attempts,retry_stream_interval_ms)
|
||||
VALUES ($1,$2,NULLIF($3,''),NULLIF($4,''),$5,NULLIF($6,''),NULLIF($7,''),$8,$9,$10,$11,$12,$13,$14,$15,$16,$17,$18,$19,$20,$21,$22,$23,$24) RETURNING id`,
|
||||
err := d.QueryRow(`INSERT INTO llm_profiles(name,format,base_url,proxy,model,api_key,api_key_hint,rate_per_second,rate_per_minute,context_window_k,reasoning_effort,priority,pool_exclude,thinking_type,streaming,max_tokens,max_tokens_field,session_header_key,retry_connect_attempts,retry_connect_interval_ms,retry_empty_attempts,retry_empty_interval_ms,retry_stream_attempts,retry_stream_interval_ms,user_agent)
|
||||
VALUES ($1,$2,NULLIF($3,''),NULLIF($4,''),$5,NULLIF($6,''),NULLIF($7,''),$8,$9,$10,$11,$12,$13,$14,$15,$16,$17,$18,$19,$20,$21,$22,$23,$24,$25) RETURNING id`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.SessionHeaderKey,
|
||||
r.Connect.Attempts, r.Connect.IntervalMS, r.Empty.Attempts, r.Empty.IntervalMS, r.Stream.Attempts, r.Stream.IntervalMS).Scan(&id)
|
||||
r.Connect.Attempts, r.Connect.IntervalMS, r.Empty.Attempts, r.Empty.IntervalMS, r.Stream.Attempts, r.Stream.IntervalMS, p.UserAgent).Scan(&id)
|
||||
if err == nil {
|
||||
_, err = d.saveProfilePrices(id, p)
|
||||
}
|
||||
return id, err
|
||||
}
|
||||
if p.APIKey == "" {
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,rate_per_second=$6,rate_per_minute=$7,context_window_k=$8,reasoning_effort=$9,priority=$10,pool_exclude=$11,thinking_type=$12,streaming=$13,max_tokens=$14,max_tokens_field=$15,session_header_key=$16,retry_connect_attempts=$17,retry_connect_interval_ms=$18,retry_empty_attempts=$19,retry_empty_interval_ms=$20,retry_stream_attempts=$21,retry_stream_interval_ms=$22 WHERE id=$23`,
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,rate_per_second=$6,rate_per_minute=$7,context_window_k=$8,reasoning_effort=$9,priority=$10,pool_exclude=$11,thinking_type=$12,streaming=$13,max_tokens=$14,max_tokens_field=$15,session_header_key=$16,retry_connect_attempts=$17,retry_connect_interval_ms=$18,retry_empty_attempts=$19,retry_empty_interval_ms=$20,retry_stream_attempts=$21,retry_stream_interval_ms=$22,user_agent=$23 WHERE id=$24`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.SessionHeaderKey,
|
||||
r.Connect.Attempts, r.Connect.IntervalMS, r.Empty.Attempts, r.Empty.IntervalMS, r.Stream.Attempts, r.Stream.IntervalMS, p.ID)
|
||||
r.Connect.Attempts, r.Connect.IntervalMS, r.Empty.Attempts, r.Empty.IntervalMS, r.Stream.Attempts, r.Stream.IntervalMS, p.UserAgent, p.ID)
|
||||
if err == nil {
|
||||
_, err = d.saveProfilePrices(p.ID, p)
|
||||
}
|
||||
return p.ID, err
|
||||
}
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,api_key=$6,api_key_hint=$7,rate_per_second=$8,rate_per_minute=$9,context_window_k=$10,reasoning_effort=$11,priority=$12,pool_exclude=$13,thinking_type=$14,streaming=$15,max_tokens=$16,max_tokens_field=$17,session_header_key=$18,retry_connect_attempts=$19,retry_connect_interval_ms=$20,retry_empty_attempts=$21,retry_empty_interval_ms=$22,retry_stream_attempts=$23,retry_stream_interval_ms=$24 WHERE id=$25`,
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,api_key=$6,api_key_hint=$7,rate_per_second=$8,rate_per_minute=$9,context_window_k=$10,reasoning_effort=$11,priority=$12,pool_exclude=$13,thinking_type=$14,streaming=$15,max_tokens=$16,max_tokens_field=$17,session_header_key=$18,retry_connect_attempts=$19,retry_connect_interval_ms=$20,retry_empty_attempts=$21,retry_empty_interval_ms=$22,retry_stream_attempts=$23,retry_stream_interval_ms=$24,user_agent=$25 WHERE id=$26`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.SessionHeaderKey,
|
||||
r.Connect.Attempts, r.Connect.IntervalMS, r.Empty.Attempts, r.Empty.IntervalMS, r.Stream.Attempts, r.Stream.IntervalMS, p.ID)
|
||||
r.Connect.Attempts, r.Connect.IntervalMS, r.Empty.Attempts, r.Empty.IntervalMS, r.Stream.Attempts, r.Stream.IntervalMS, p.UserAgent, p.ID)
|
||||
if err == nil {
|
||||
_, err = d.saveProfilePrices(p.ID, p)
|
||||
}
|
||||
|
||||
@@ -396,6 +396,8 @@ CREATE TABLE IF NOT EXISTS llm_profiles (
|
||||
-- Custom session header: when non-empty, every request carries an HTTP header of this name whose value is the current session id
|
||||
-- (chat session/worker intent). For gateways that do prompt caching/sticky routing by a session-id header. ''=do not send.
|
||||
session_header_key TEXT NOT NULL DEFAULT '',
|
||||
-- Custom User-Agent: when non-empty, every outbound LLM request overrides the User-Agent header with this value. ''=do not override (keep the default UA).
|
||||
user_agent TEXT NOT NULL DEFAULT '',
|
||||
-- Retry overrides: attempts 0=use global default / -1=disable / >0=use this value; interval 0=use default exponential backoff / >0=fixed milliseconds.
|
||||
-- The three groups correspond to connect retry, empty-response retry, and same-provider safe-window retry; see the ALTER comment below.
|
||||
retry_connect_attempts INTEGER NOT NULL DEFAULT 0,
|
||||
@@ -436,6 +438,8 @@ ALTER TABLE llm_profiles ADD CONSTRAINT llm_profiles_max_tokens_check
|
||||
CHECK (max_tokens >= 0);
|
||||
-- Custom session header name; backfills old databases. Default '' = do not send; old configs behave unchanged.
|
||||
ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS session_header_key TEXT NOT NULL DEFAULT '';
|
||||
-- Custom User-Agent; backfills old databases. Default '' = do not override (keep the default UA); old configs behave unchanged.
|
||||
ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS user_agent TEXT NOT NULL DEFAULT '';
|
||||
|
||||
-- Per-profile retry overrides (see the LLM retry design doc). Each of the three groups is a pair of
|
||||
-- "attempts + fixed interval" with unified semantics: attempts 0=use the global default, -1=disable this layer's
|
||||
|
||||
@@ -617,6 +617,9 @@ type LLMProfileDTO struct {
|
||||
// Custom session header name: when non-empty, every request carries this HTTP header,
|
||||
// its value = the current session/intent's session id. '' = do not send. For gateways that key prompt caching/sticky routing off a session-id header.
|
||||
SessionHeaderKey string `json:"session_header_key"`
|
||||
// Custom User-Agent: when non-empty, every outbound LLM request overrides its
|
||||
// User-Agent header with this value. '' = do not override (keep the default UA).
|
||||
UserAgent string `json:"user_agent"`
|
||||
// This profile's retry overrides (connect / empty response / same-provider safe
|
||||
// window). Per item attempts: 0=inherit the global policy | -1=disable that retry
|
||||
// layer | >0=count; interval_ms: 0=default exponential backoff | >0=use that fixed millisecond interval. All zero = fully follow the global policy, i.e. the historical behavior.
|
||||
@@ -644,6 +647,7 @@ func llmProfileDTO(p *db.LLMProfile) LLMProfileDTO {
|
||||
MaxTokens: p.MaxTokens,
|
||||
MaxTokensField: p.MaxTokensField,
|
||||
SessionHeaderKey: p.SessionHeaderKey,
|
||||
UserAgent: p.UserAgent,
|
||||
Retry: p.Retry,
|
||||
}
|
||||
}
|
||||
|
||||
+12
-1
@@ -355,6 +355,7 @@ func (s *Server) loadLLMConfig() (agent.Config, bool) {
|
||||
cfg.Stream = p.Streaming
|
||||
cfg.MaxTokens, cfg.MaxTokensField = p.MaxTokens, p.MaxTokensField
|
||||
cfg.SessionHeaderKey = p.SessionHeaderKey
|
||||
cfg.UserAgent = p.UserAgent
|
||||
s.applyProfileRetry(&cfg, p)
|
||||
// Ollama profiles run keyless (local server); the other formats need a key.
|
||||
if cfg.APIKey == "" && cfg.NeedsAPIKey() {
|
||||
@@ -381,6 +382,7 @@ func (s *Server) saveLLMConfig(cfg agent.Config) error {
|
||||
var maxTokens int
|
||||
var maxTokensField string
|
||||
var sessionHeaderKey string
|
||||
var userAgent string
|
||||
var retry db.RetryOverride
|
||||
if profs, _ := s.m.pg.ListProfiles(); profs != nil {
|
||||
for _, p := range profs {
|
||||
@@ -388,6 +390,7 @@ func (s *Server) saveLLMConfig(cfg agent.Config) error {
|
||||
id, priority, poolExclude, streaming = p.ID, p.Priority, p.PoolExclude, p.Streaming
|
||||
maxTokens, maxTokensField = p.MaxTokens, p.MaxTokensField
|
||||
sessionHeaderKey = p.SessionHeaderKey
|
||||
userAgent = p.UserAgent
|
||||
retry = p.Retry
|
||||
break
|
||||
}
|
||||
@@ -398,7 +401,7 @@ func (s *Server) saveLLMConfig(cfg agent.Config) error {
|
||||
APIKey: cfg.APIKey, RatePerSecond: cfg.RatePerSecond, RatePerMinute: cfg.RatePerMinute,
|
||||
ContextWindowK: cfg.ContextWindowK, ThinkingType: cfg.ThinkingType, ReasoningEffort: cfg.ReasoningEffort, IsDefault: true,
|
||||
Priority: priority, PoolExclude: poolExclude, Streaming: streaming,
|
||||
MaxTokens: maxTokens, MaxTokensField: maxTokensField, SessionHeaderKey: sessionHeaderKey,
|
||||
MaxTokens: maxTokens, MaxTokensField: maxTokensField, SessionHeaderKey: sessionHeaderKey, UserAgent: userAgent,
|
||||
Retry: retry,
|
||||
})
|
||||
if err != nil {
|
||||
@@ -528,6 +531,7 @@ func (s *Server) loadProfileConfig(id int64) (agent.Config, bool) {
|
||||
cfg.Stream = p.Streaming
|
||||
cfg.MaxTokens, cfg.MaxTokensField = p.MaxTokens, p.MaxTokensField
|
||||
cfg.SessionHeaderKey = p.SessionHeaderKey
|
||||
cfg.UserAgent = p.UserAgent
|
||||
s.applyProfileRetry(&cfg, p)
|
||||
if cfg.APIKey == "" && cfg.NeedsAPIKey() {
|
||||
return cfg, false
|
||||
@@ -1418,6 +1422,7 @@ func (s *Server) testLLM(w http.ResponseWriter, r *http.Request) {
|
||||
ProfileID *int64 `json:"profile_id"` // passed when testing a stored profile: if api_key is empty, its stored key is used
|
||||
Streaming *bool `json:"streaming"` // omitted = streaming, the same defaults as when saving a profile
|
||||
SessionHeaderKey string `json:"session_header_key"` // non-empty = the test also sends this custom session header, with a one-off random session id as the value
|
||||
UserAgent string `json:"user_agent"` // non-empty = the test overrides the User-Agent header with this value (mirrors production)
|
||||
}
|
||||
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
|
||||
writeErr(w, 400, err.Error())
|
||||
@@ -1477,6 +1482,12 @@ func (s *Server) testLLM(w http.ResponseWriter, r *http.Request) {
|
||||
if stored != nil && cfg.SessionHeaderKey == "" {
|
||||
cfg.SessionHeaderKey = strings.TrimSpace(stored.SessionHeaderKey)
|
||||
}
|
||||
// The custom User-Agent follows the profile too, so the test request uses the
|
||||
// exact UA a real conversation would; the form value wins, else the stored one.
|
||||
cfg.UserAgent = strings.TrimSpace(req.UserAgent)
|
||||
if stored != nil && cfg.UserAgent == "" {
|
||||
cfg.UserAgent = strings.TrimSpace(stored.UserAgent)
|
||||
}
|
||||
// Keyless formats (Ollama) skip the key resolution entirely.
|
||||
if cfg.NeedsAPIKey() {
|
||||
// API key resolution precedence: form input > the named profile's stored key > the global config's key.
|
||||
|
||||
@@ -1882,6 +1882,14 @@ func (s *Server) pgSaveProfile(w http.ResponseWriter, r *http.Request) {
|
||||
if p.Format != "openai" || p.MaxTokensField != llm.MaxTokensFieldCompletion {
|
||||
p.MaxTokensField = ""
|
||||
}
|
||||
// Custom User-Agent: trim and reject control characters / over-long values so a
|
||||
// bad value can never inject a header or bloat every request. "" = no override.
|
||||
ua, err := agent.ValidateUserAgent(p.UserAgent)
|
||||
if err != nil {
|
||||
writeErr(w, 400, err.Error())
|
||||
return
|
||||
}
|
||||
p.UserAgent = ua
|
||||
id, err := pg.SaveProfile(&p)
|
||||
if err != nil {
|
||||
writeErr(w, 500, err.Error())
|
||||
|
||||
@@ -41,7 +41,7 @@ type sideQuestionState struct {
|
||||
}
|
||||
|
||||
func sideModel(cfg agent.Config, id int64, name string) sidequestion.Model {
|
||||
b, _ := json.Marshal([]any{cfg.Format, cfg.BaseURL, cfg.Model, cfg.ThinkingType, cfg.ReasoningEffort, cfg.Stream, cfg.MaxTokensField, cfg.SessionHeaderKey})
|
||||
b, _ := json.Marshal([]any{cfg.Format, cfg.BaseURL, cfg.Model, cfg.ThinkingType, cfg.ReasoningEffort, cfg.Stream, cfg.MaxTokensField, cfg.SessionHeaderKey, cfg.UserAgent})
|
||||
h := sha256.Sum256(b)
|
||||
return sidequestion.Model{ProfileID: id, Name: name, Format: string(cfg.Format), Model: cfg.Model, Identity: hex.EncodeToString(h[:]), Streaming: cfg.Stream, WindowTokens: cfg.CompactionWindow()}
|
||||
}
|
||||
|
||||
@@ -335,6 +335,7 @@ function ProfileSheet({
|
||||
const [maxTokens, setMaxTokens] = React.useState("0"); // per-reply output cap; 0 = not sent
|
||||
const [maxTokensField, setMaxTokensField] = React.useState(NONE); // which field name carries the cap; NONE = max_tokens
|
||||
const [sessionHeaderKey, setSessionHeaderKey] = React.useState(""); // custom session header name; empty = not sent
|
||||
const [userAgent, setUserAgent] = React.useState(""); // custom User-Agent; empty = keep the default UA
|
||||
const [retry, setRetry] = React.useState<LLMRetryOverride>(ZERO_OVERRIDE); // retry override for this profile; all 0 = follow the global
|
||||
const [testing, setTesting] = React.useState(false);
|
||||
const [saving, setSaving] = React.useState(false);
|
||||
@@ -366,6 +367,7 @@ function ProfileSheet({
|
||||
setMaxTokens(String(profile?.max_tokens ?? 0));
|
||||
setMaxTokensField(fromStore(profile?.max_tokens_field));
|
||||
setSessionHeaderKey(profile?.session_header_key ?? "");
|
||||
setUserAgent(profile?.user_agent ?? "");
|
||||
setRetry(profile?.retry ?? ZERO_OVERRIDE);
|
||||
setApiKey("");
|
||||
setKeyHint(profile?.api_key_hint ?? "");
|
||||
@@ -412,6 +414,7 @@ function ProfileSheet({
|
||||
profileId,
|
||||
streaming,
|
||||
sessionHeaderKey.trim(),
|
||||
userAgent.trim(),
|
||||
);
|
||||
// Show the reply too: only when the model demonstrably spoke is this the same as working inside a session.
|
||||
if (r.ok)
|
||||
@@ -455,6 +458,7 @@ function ProfileSheet({
|
||||
// and the backend re-applies the same normalization; this just keeps the UI from sending contradictory values.
|
||||
max_tokens_field: format === "openai" ? toStore(maxTokensField) : "",
|
||||
session_header_key: sessionHeaderKey.trim(),
|
||||
user_agent: userAgent.trim(),
|
||||
retry,
|
||||
});
|
||||
if (isNew) toast.success(`Created: ${name.trim()} (use "Set as active" on its card to enable it)`);
|
||||
@@ -613,6 +617,21 @@ function ProfileSheet({
|
||||
</p>
|
||||
</div>
|
||||
|
||||
<div className="grid gap-2">
|
||||
<Label htmlFor="p-user-agent">Custom User-Agent (optional)</Label>
|
||||
<Input
|
||||
id="p-user-agent"
|
||||
className="font-mono"
|
||||
placeholder="e.g. my-app/1.0 (empty = default UA)"
|
||||
value={userAgent}
|
||||
onChange={(e) => setUserAgent(e.target.value)}
|
||||
/>
|
||||
<p className="text-muted-foreground text-xs">
|
||||
Overrides the <b>User-Agent</b> header on every outbound LLM request. Useful for gateways that route or allow-list by
|
||||
User-Agent. Leave empty to keep the default User-Agent unchanged.
|
||||
</p>
|
||||
</div>
|
||||
|
||||
<div className="grid gap-2">
|
||||
<Label htmlFor="p-api-key">
|
||||
{format === "ollama" ? "API Key (optional for Ollama)" : "API Key"}
|
||||
|
||||
@@ -949,6 +949,7 @@ export const api = {
|
||||
profile_id?: number,
|
||||
streaming = true, // test with the profile's real streaming mode so "streaming works, non-streaming does not" never leaks into conversations
|
||||
session_header_key = "", // non-empty = the test request also carries this custom session header (value is a one-off session id)
|
||||
user_agent = "", // non-empty = the test overrides the User-Agent header with this value
|
||||
) =>
|
||||
// reply = the model's actual reply (truncated); the backend marks a profile that replies with nothing as failed
|
||||
post<{ ok: boolean; error?: string; latency_ms?: number; model?: string; reply?: string }>("/llm/test", {
|
||||
@@ -962,6 +963,7 @@ export const api = {
|
||||
profile_id,
|
||||
streaming,
|
||||
session_header_key,
|
||||
user_agent,
|
||||
}),
|
||||
llmProfiles: () => get<{ profiles: LLMProfile[] }>("/llm/profiles").then((r) => arr(r.profiles)),
|
||||
saveLLMProfile: (p: {
|
||||
@@ -983,6 +985,7 @@ export const api = {
|
||||
max_tokens?: number; // output cap per reply; 0 = not sent, the server default decides
|
||||
max_tokens_field?: string; // ""=max_tokens (default) | "max_completion_tokens" (openai format only)
|
||||
session_header_key?: string; // non-empty = send this HTTP header on every request, value = the current session id; "" = not sent
|
||||
user_agent?: string; // non-empty = override the User-Agent header on every request; "" = keep the default UA
|
||||
retry?: LLMRetryOverride; // this profile's retry override; all zeros = follow the global retry policy
|
||||
}) => post<{ id: number }>("/llm/profiles", p),
|
||||
deleteLLMProfile: (id: string) => del<{ deleted: number }>(`/llm/profiles/${id}`),
|
||||
|
||||
@@ -1061,6 +1061,8 @@ export interface LLMProfile {
|
||||
// Custom session header name: when non-empty every request carries this HTTP header, value = the current session/intent session id.
|
||||
// "" = not sent. For gateways that do prompt caching/sticky routing on a session-id header.
|
||||
session_header_key?: string;
|
||||
// Custom User-Agent: when non-empty, every outbound LLM request overrides its User-Agent header with this value. "" = do not override (keep the default UA).
|
||||
user_agent?: string;
|
||||
// This profile's retry overrides (connect / empty response / same-provider safety window). Empty / all zeros = follow the global policy.
|
||||
retry?: LLMRetryOverride;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user