feat: Plexum-kimi-provider v0.1 — Moonshot Kimi via coding endpoint
Plexum ProviderPlugin that serves Moonshot Kimi K2.6 ("Kimi for
Coding") via Kimi's Anthropic-compatible coding endpoint at
https://api.kimi.com/coding/v1/messages. Port of openclaw's
extensions/kimi-coding/provider-catalog.ts to Go + Plexum SDK.
Repo structure parallels Plexum-minimax-provider:
- internal/anthropic/ HTTP+SSE Anthropic Messages client, with new
UserAgent field (Kimi expects "claude-code/
0.1.0" — openclaw plugin parity)
- internal/translate/ canonical ↔ Anthropic translator (re-used
shape from MiniMax — no Kimi-specific quirks
needed for v1 plain-text path)
- cmd/plexum-kimi-provider-plugin/ ProviderPlugin entry
Declared models (Kimi server accepts all three; plugin normalizes
legacy aliases to the canonical id on the wire):
kimi-for-coding (current, default)
kimi-code (legacy alias)
k2p5 (legacy alias)
HostConfig: api_key (required), base_url (override), user_agent
(default "claude-code/0.1.0"), max_tokens_default (default 8192).
End-to-end verified against the live `sk-kimi-` subscription key:
1. CLI embedded turn 1: "Hi there! I'm Kimi."
2. CLI embedded turn 2: "I said hi, I'm Kimi." (multi-turn context OK)
3. Via gateway socket: {"outcome":"text","text":"...pong"}
4. Via Fabric channel: alice → bt2-clean → kimi agent → Kimi K2.6 →
outbound REST → reply in channel seq=15:
"Concise concurrency: goroutines and channels
make parallel code readable, safe, and
efficient without the usual threading
complexity."
Test the bidirectional channel pipeline works with a fresh provider:
Fabric (channel plugin) + Kimi (provider plugin) wired through Plexum
agentloop, MiniMax-style plugin packaging.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
79
README.md
Normal file
79
README.md
Normal file
@@ -0,0 +1,79 @@
|
||||
# Plexum-kimi-provider
|
||||
|
||||
Plexum ProviderPlugin that serves **Moonshot Kimi K2.6 ("Kimi for Coding")**
|
||||
via Kimi's Anthropic-compatible coding endpoint
|
||||
(`https://api.kimi.com/coding/v1/messages`).
|
||||
|
||||
Port of `openclaw/extensions/kimi-coding/provider-catalog.ts` to Go +
|
||||
Plexum SDK.
|
||||
|
||||
## Status
|
||||
|
||||
**v0.1 — current**: API key auth (Kimi subscription `sk-kimi-...`),
|
||||
streaming SSE, models `kimi-for-coding` / `kimi-code` / `k2p5`. Sends
|
||||
`User-Agent: claude-code/0.1.0` (mirrors openclaw plugin) for parity
|
||||
with Kimi's coding subscription auth path.
|
||||
|
||||
## Install
|
||||
|
||||
```bash
|
||||
cd ~/Plexum-kimi-provider
|
||||
./scripts/install.sh
|
||||
```
|
||||
|
||||
Then:
|
||||
|
||||
1. **Write API key** to `~/.plexum/plugins/plexum-kimi-provider/config.json`:
|
||||
```json
|
||||
{"api_key": "sk-kimi-..."}
|
||||
```
|
||||
(`chmod 600` it.)
|
||||
|
||||
2. **Allow the plugin** in `~/.plexum/plexum.json`:
|
||||
```json
|
||||
{"plugins": {"allow": ["plexum-kimi-provider"]}}
|
||||
```
|
||||
|
||||
3. **Point an agent at Kimi**:
|
||||
```bash
|
||||
plexum agent-add --model kimi-for-coding my-agent
|
||||
```
|
||||
|
||||
4. **Restart** and talk:
|
||||
```bash
|
||||
systemctl --user restart plexum
|
||||
plexum say --agent-id my-agent --session-id $(plexum new-session --agent-id my-agent) "hello"
|
||||
```
|
||||
|
||||
## Config options
|
||||
|
||||
| Field | Default | Notes |
|
||||
|---|---|---|
|
||||
| `api_key` | (required) | `sk-kimi-...` subscription key |
|
||||
| `base_url` | `https://api.kimi.com/coding` | override only if you proxy |
|
||||
| `user_agent` | `claude-code/0.1.0` | matches openclaw plugin |
|
||||
| `max_tokens_default` | `8192` | used when TurnRequest.MaxTokens unset |
|
||||
|
||||
## Model id aliases
|
||||
|
||||
Kimi server accepts all three; the plugin normalizes to
|
||||
`kimi-for-coding` on the wire for consistent logs:
|
||||
|
||||
| advertised id | wire id |
|
||||
|---|---|
|
||||
| `kimi-for-coding` | `kimi-for-coding` |
|
||||
| `kimi-code` (legacy) | `kimi-for-coding` |
|
||||
| `k2p5` (legacy) | `kimi-for-coding` |
|
||||
|
||||
## Architecture
|
||||
|
||||
Same shape as `Plexum-minimax-provider`:
|
||||
|
||||
- `internal/anthropic/` — HTTP+SSE Messages client (now with `UserAgent`
|
||||
field; can be shared by future Anthropic-compat providers)
|
||||
- `internal/translate/` — canonical ↔ Anthropic translator
|
||||
- `cmd/plexum-kimi-provider-plugin/` — ProviderPlugin entry
|
||||
|
||||
## License
|
||||
|
||||
Same as Plexum.
|
||||
169
cmd/plexum-kimi-provider-plugin/main.go
Normal file
169
cmd/plexum-kimi-provider-plugin/main.go
Normal file
@@ -0,0 +1,169 @@
|
||||
// plexum-kimi-provider-plugin is a Plexum ProviderPlugin that serves
|
||||
// Moonshot Kimi K2.6 ("Kimi for Coding") via Kimi's Anthropic-compatible
|
||||
// coding endpoint (https://api.kimi.com/coding/v1/messages).
|
||||
//
|
||||
// Reference: openclaw/extensions/kimi-coding/provider-catalog.ts —
|
||||
// same base URL + User-Agent convention. The endpoint accepts an
|
||||
// `sk-kimi-` subscription API key minted from kimi.com/code.
|
||||
//
|
||||
// Declared models:
|
||||
// - kimi-for-coding (current; default)
|
||||
// - kimi-code (legacy alias; same model server-side)
|
||||
// - k2p5 (legacy alias; same)
|
||||
//
|
||||
// Operator agent.json.model = "kimi-for-coding" routes here via
|
||||
// Plexum's ProviderRouter.
|
||||
package main
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
|
||||
"git.hangman-lab.top/hzhang/Plexum-sdk-go/canonical"
|
||||
plugin "git.hangman-lab.top/hzhang/Plexum-sdk-go/plugin"
|
||||
|
||||
"git.hangman-lab.top/hzhang/Plexum-kimi-provider/internal/anthropic"
|
||||
"git.hangman-lab.top/hzhang/Plexum-kimi-provider/internal/translate"
|
||||
)
|
||||
|
||||
const (
|
||||
pluginName = "plexum-kimi-provider"
|
||||
|
||||
// Kimi coding endpoint. The path /v1/messages is appended by the
|
||||
// anthropic client; baseURL must NOT include /v1.
|
||||
defaultBaseURL = "https://api.kimi.com/coding"
|
||||
|
||||
// UA that openclaw's kimi-coding plugin sets. Likely a server-side
|
||||
// allowlist marker for the coding subscription auth path; omitting
|
||||
// it currently still works but we send it for parity.
|
||||
defaultUserAgent = "claude-code/0.1.0"
|
||||
|
||||
// Kimi K2.6 supports up to 32768 output tokens per request.
|
||||
defaultMaxTokens = 8192
|
||||
)
|
||||
|
||||
// supportedModels = what manifest.contracts.provider.models advertises.
|
||||
// The legacy aliases (kimi-code, k2p5) route the same way; Kimi server
|
||||
// accepts them and returns kimi-for-coding-equivalent output.
|
||||
var supportedModels = []string{"kimi-for-coding", "kimi-code", "k2p5"}
|
||||
|
||||
// HostConfig is per-profile plugin config at
|
||||
// <profile>/plugins/plexum-kimi-provider/config.json:
|
||||
//
|
||||
// {
|
||||
// "api_key": "sk-kimi-...", // required
|
||||
// "base_url": "https://...", // optional override
|
||||
// "user_agent": "...", // optional; default "claude-code/0.1.0"
|
||||
// "max_tokens_default": 8192 // optional
|
||||
// }
|
||||
type HostConfig struct {
|
||||
APIKey string `json:"api_key"`
|
||||
BaseURL string `json:"base_url"`
|
||||
UserAgent string `json:"user_agent"`
|
||||
MaxTokensDefault int `json:"max_tokens_default"`
|
||||
}
|
||||
|
||||
type kimiPlugin struct {
|
||||
host plugin.HostAPI
|
||||
cfg HostConfig
|
||||
cli *anthropic.Client
|
||||
}
|
||||
|
||||
func (p *kimiPlugin) Manifest() plugin.Manifest {
|
||||
return plugin.Manifest{
|
||||
Name: pluginName,
|
||||
Version: "0.1.0",
|
||||
Activation: plugin.ActivationLazy,
|
||||
Executable: "plexum-kimi-provider-plugin",
|
||||
Contracts: plugin.Contracts{
|
||||
Provider: &plugin.ProviderContract{Models: supportedModels},
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
func (p *kimiPlugin) Init(ctx context.Context, host plugin.HostAPI) error {
|
||||
p.host = host
|
||||
|
||||
profileRoot := os.Getenv("PLEXUM_PROFILE_ROOT")
|
||||
if profileRoot == "" {
|
||||
home, _ := os.UserHomeDir()
|
||||
profileRoot = filepath.Join(home, ".plexum")
|
||||
}
|
||||
cfgPath := filepath.Join(profileRoot, "plugins", pluginName, "config.json")
|
||||
raw, err := os.ReadFile(cfgPath)
|
||||
if err != nil && !errors.Is(err, os.ErrNotExist) {
|
||||
return fmt.Errorf("read %s: %w", cfgPath, err)
|
||||
}
|
||||
if len(raw) > 0 {
|
||||
if err := json.Unmarshal(raw, &p.cfg); err != nil {
|
||||
return fmt.Errorf("parse %s: %w", cfgPath, err)
|
||||
}
|
||||
}
|
||||
if p.cfg.APIKey == "" {
|
||||
return fmt.Errorf("kimi: api_key missing in %s", cfgPath)
|
||||
}
|
||||
base := p.cfg.BaseURL
|
||||
if base == "" {
|
||||
base = defaultBaseURL
|
||||
}
|
||||
ua := p.cfg.UserAgent
|
||||
if ua == "" {
|
||||
ua = defaultUserAgent
|
||||
}
|
||||
if p.cfg.MaxTokensDefault <= 0 {
|
||||
p.cfg.MaxTokensDefault = defaultMaxTokens
|
||||
}
|
||||
p.cli = anthropic.New(base, p.cfg.APIKey)
|
||||
p.cli.UserAgent = ua
|
||||
host.Log("info", "kimi provider initialized", map[string]any{
|
||||
"base": base, "user_agent": ua, "models": supportedModels,
|
||||
"max_tokens_default": p.cfg.MaxTokensDefault,
|
||||
})
|
||||
return nil
|
||||
}
|
||||
|
||||
// Stream is the ProviderPlugin entrypoint. canonical.TurnRequest in,
|
||||
// channel of canonical.TurnEvent out.
|
||||
func (p *kimiPlugin) Stream(ctx context.Context, modelID string, req canonical.TurnRequest) (<-chan canonical.TurnEvent, error) {
|
||||
// Normalize legacy aliases to the current canonical id so the wire
|
||||
// model name is consistent (Kimi server accepts all 3, but this
|
||||
// keeps logs + telemetry tidy).
|
||||
wireModel := modelID
|
||||
if wireModel == "kimi-code" || wireModel == "k2p5" {
|
||||
wireModel = "kimi-for-coding"
|
||||
}
|
||||
apiReq, err := translate.CanonicalToAnthropic(req, wireModel, p.cfg.MaxTokensDefault)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
raw, err := p.cli.StreamMessages(ctx, apiReq)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
out := make(chan canonical.TurnEvent, 32)
|
||||
go func() {
|
||||
defer close(out)
|
||||
tr := translate.NewTranslator()
|
||||
for ev := range raw {
|
||||
for _, te := range tr.Translate(ev) {
|
||||
select {
|
||||
case out <- te:
|
||||
case <-ctx.Done():
|
||||
return
|
||||
}
|
||||
}
|
||||
}
|
||||
}()
|
||||
return out, nil
|
||||
}
|
||||
|
||||
func main() {
|
||||
if err := plugin.Serve(&kimiPlugin{}); err != nil {
|
||||
fmt.Fprintf(os.Stderr, "plexum-kimi-provider-plugin: %v\n", err)
|
||||
os.Exit(1)
|
||||
}
|
||||
}
|
||||
7
go.mod
Normal file
7
go.mod
Normal file
@@ -0,0 +1,7 @@
|
||||
module git.hangman-lab.top/hzhang/Plexum-kimi-provider
|
||||
|
||||
go 1.24.2
|
||||
|
||||
require git.hangman-lab.top/hzhang/Plexum-sdk-go v0.0.0
|
||||
|
||||
replace git.hangman-lab.top/hzhang/Plexum-sdk-go => ../Plexum-sdk-go
|
||||
278
internal/anthropic/client.go
Normal file
278
internal/anthropic/client.go
Normal file
@@ -0,0 +1,278 @@
|
||||
// Package anthropic is a minimal Anthropic Messages API HTTP+SSE
|
||||
// client. Works against the real Anthropic API and any Anthropic-
|
||||
// compatible endpoint (MiniMax exposes one at
|
||||
// https://api.minimax.io/anthropic).
|
||||
//
|
||||
// The MiniMax provider plugin's only consumer; if other providers
|
||||
// land later they can reuse this package.
|
||||
package anthropic
|
||||
|
||||
import (
|
||||
"bufio"
|
||||
"bytes"
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"io"
|
||||
"net/http"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
// DefaultAPIVersion matches what Anthropic + Anthropic-compat servers expect.
|
||||
const DefaultAPIVersion = "2023-06-01"
|
||||
|
||||
// Client is a thin HTTP wrapper. Stateless across calls.
|
||||
type Client struct {
|
||||
BaseURL string // e.g. "https://api.kimi.com/coding" or "https://api.minimax.io/anthropic"
|
||||
APIKey string
|
||||
APIVersion string // default "2023-06-01"
|
||||
// UserAgent, if set, replaces Go's default UA header. Kimi's
|
||||
// coding endpoint expects "claude-code/0.1.0"; other backends
|
||||
// (MiniMax etc.) are flexible.
|
||||
UserAgent string
|
||||
// ExtraHeaders are sent verbatim on every request.
|
||||
ExtraHeaders map[string]string
|
||||
HTTP *http.Client
|
||||
}
|
||||
|
||||
// New constructs a Client with sensible defaults.
|
||||
func New(baseURL, apiKey string) *Client {
|
||||
return &Client{
|
||||
BaseURL: strings.TrimRight(baseURL, "/"),
|
||||
APIKey: apiKey,
|
||||
APIVersion: DefaultAPIVersion,
|
||||
// 5 min ceiling per turn — long enough for a slow reasoning
|
||||
// response, short enough that wedged calls don't hang forever.
|
||||
HTTP: &http.Client{Timeout: 5 * time.Minute},
|
||||
}
|
||||
}
|
||||
|
||||
// MessagesRequest is the wire shape POSTed to /v1/messages.
|
||||
type MessagesRequest struct {
|
||||
Model string `json:"model"`
|
||||
System any `json:"system,omitempty"` // string OR []ContentBlock
|
||||
Messages []Message `json:"messages"`
|
||||
MaxTokens int `json:"max_tokens"`
|
||||
Temperature *float64 `json:"temperature,omitempty"`
|
||||
StopSequences []string `json:"stop_sequences,omitempty"`
|
||||
Tools []ToolDef `json:"tools,omitempty"`
|
||||
ToolChoice *ToolChoice `json:"tool_choice,omitempty"`
|
||||
Thinking *ThinkingConfig `json:"thinking,omitempty"`
|
||||
Stream bool `json:"stream"`
|
||||
Metadata *RequestMetadata `json:"metadata,omitempty"`
|
||||
}
|
||||
|
||||
// Message is one entry in messages[].
|
||||
type Message struct {
|
||||
Role string `json:"role"` // "user" | "assistant"
|
||||
Content []ContentBlock `json:"content"`
|
||||
}
|
||||
|
||||
// ContentBlock covers text / tool_use / tool_result / thinking. We
|
||||
// keep the discriminated union loose (map[string]any) on the request
|
||||
// side so callers can pass whatever shape the API accepts; on the
|
||||
// response side the SSE parser produces typed deltas.
|
||||
type ContentBlock = map[string]any
|
||||
|
||||
// ToolDef is one tool entry the model can call.
|
||||
type ToolDef struct {
|
||||
Name string `json:"name"`
|
||||
Description string `json:"description,omitempty"`
|
||||
InputSchema json.RawMessage `json:"input_schema"`
|
||||
}
|
||||
|
||||
// ToolChoice constrains how the model picks tools.
|
||||
type ToolChoice struct {
|
||||
Type string `json:"type"` // "auto" | "any" | "tool" | "none"
|
||||
Name string `json:"name,omitempty"`
|
||||
}
|
||||
|
||||
// ThinkingConfig enables extended-thinking on supported models.
|
||||
type ThinkingConfig struct {
|
||||
Type string `json:"type"` // "enabled" | "disabled"
|
||||
BudgetTokens int `json:"budget_tokens,omitempty"`
|
||||
}
|
||||
|
||||
// RequestMetadata is optional user-supplied tracing.
|
||||
type RequestMetadata struct {
|
||||
UserID string `json:"user_id,omitempty"`
|
||||
}
|
||||
|
||||
// ---- Streaming SSE response ----
|
||||
|
||||
// Event is one parsed SSE event from the streaming endpoint.
|
||||
// Direct mirror of Anthropic's event shapes — caller's responsibility
|
||||
// to translate into Plexum's canonical.TurnEvent.
|
||||
type Event struct {
|
||||
Type string `json:"type"`
|
||||
// message_start: complete initial message envelope
|
||||
Message *MessageEnvelope `json:"message,omitempty"`
|
||||
// content_block_start: block at .Index
|
||||
Index int `json:"index,omitempty"`
|
||||
ContentBlock *BlockStart `json:"content_block,omitempty"`
|
||||
// content_block_delta: text_delta / thinking_delta / input_json_delta / signature_delta
|
||||
Delta *BlockDelta `json:"delta,omitempty"`
|
||||
// message_delta: usage update + stop_reason
|
||||
Usage *Usage `json:"usage,omitempty"`
|
||||
// error event payload
|
||||
Error *ErrorBody `json:"error,omitempty"`
|
||||
}
|
||||
|
||||
// MessageEnvelope is the top-level message info in message_start.
|
||||
type MessageEnvelope struct {
|
||||
ID string `json:"id"`
|
||||
Type string `json:"type"`
|
||||
Role string `json:"role"`
|
||||
Model string `json:"model"`
|
||||
StopReason string `json:"stop_reason"`
|
||||
Usage *Usage `json:"usage,omitempty"`
|
||||
}
|
||||
|
||||
// BlockStart describes a block that just started. Type is "text",
|
||||
// "thinking", or "tool_use". For tool_use the ID + Name are set.
|
||||
type BlockStart struct {
|
||||
Type string `json:"type"`
|
||||
Text string `json:"text,omitempty"`
|
||||
Thinking string `json:"thinking,omitempty"`
|
||||
ID string `json:"id,omitempty"`
|
||||
Name string `json:"name,omitempty"`
|
||||
Input json.RawMessage `json:"input,omitempty"`
|
||||
}
|
||||
|
||||
// BlockDelta carries incremental changes. Exactly one of the *_delta
|
||||
// fields is meaningful per delta.
|
||||
type BlockDelta struct {
|
||||
Type string `json:"type"` // "text_delta" | "thinking_delta" | "input_json_delta" | "signature_delta"
|
||||
Text string `json:"text,omitempty"`
|
||||
Thinking string `json:"thinking,omitempty"`
|
||||
PartialJSON string `json:"partial_json,omitempty"`
|
||||
Signature string `json:"signature,omitempty"`
|
||||
// On message_delta event, the same Delta carries stop_reason.
|
||||
StopReason string `json:"stop_reason,omitempty"`
|
||||
StopSequence string `json:"stop_sequence,omitempty"`
|
||||
}
|
||||
|
||||
// Usage is the running input/output token counts.
|
||||
type Usage struct {
|
||||
InputTokens int `json:"input_tokens,omitempty"`
|
||||
OutputTokens int `json:"output_tokens,omitempty"`
|
||||
CacheReadInputTokens int `json:"cache_read_input_tokens,omitempty"`
|
||||
CacheCreationInputTokens int `json:"cache_creation_input_tokens,omitempty"`
|
||||
}
|
||||
|
||||
// ErrorBody is the Anthropic error envelope.
|
||||
type ErrorBody struct {
|
||||
Type string `json:"type"`
|
||||
Message string `json:"message"`
|
||||
}
|
||||
|
||||
// StreamMessages opens a streaming POST /v1/messages call and sends
|
||||
// parsed events down the returned channel. Channel closes when SSE
|
||||
// stream terminates (message_stop, server EOF, ctx cancel, or error).
|
||||
// On HTTP-level error (non-2xx, network failure), an EventType="error"
|
||||
// event lands on the channel with the wrapped error before close.
|
||||
func (c *Client) StreamMessages(ctx context.Context, req MessagesRequest) (<-chan Event, error) {
|
||||
req.Stream = true
|
||||
body, err := json.Marshal(req)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("anthropic: marshal: %w", err)
|
||||
}
|
||||
httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost,
|
||||
c.BaseURL+"/v1/messages", bytes.NewReader(body))
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
httpReq.Header.Set("content-type", "application/json")
|
||||
httpReq.Header.Set("authorization", "Bearer "+c.APIKey)
|
||||
httpReq.Header.Set("anthropic-version", c.APIVersion)
|
||||
httpReq.Header.Set("accept", "text/event-stream")
|
||||
if c.UserAgent != "" {
|
||||
httpReq.Header.Set("User-Agent", c.UserAgent)
|
||||
}
|
||||
for k, v := range c.ExtraHeaders {
|
||||
httpReq.Header.Set(k, v)
|
||||
}
|
||||
|
||||
resp, err := c.HTTP.Do(httpReq)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("anthropic: %w", err)
|
||||
}
|
||||
if resp.StatusCode < 200 || resp.StatusCode >= 300 {
|
||||
raw, _ := io.ReadAll(resp.Body)
|
||||
resp.Body.Close()
|
||||
return nil, fmt.Errorf("anthropic: %s -> %d: %s", c.BaseURL, resp.StatusCode, string(raw))
|
||||
}
|
||||
|
||||
ch := make(chan Event, 32)
|
||||
go func() {
|
||||
defer close(ch)
|
||||
defer resp.Body.Close()
|
||||
if err := parseSSE(ctx, resp.Body, ch); err != nil && !errors.Is(err, io.EOF) {
|
||||
select {
|
||||
case ch <- Event{Type: "error", Error: &ErrorBody{Type: "stream_error", Message: err.Error()}}:
|
||||
default:
|
||||
}
|
||||
}
|
||||
}()
|
||||
return ch, nil
|
||||
}
|
||||
|
||||
// parseSSE reads SSE frames (`event:` + `data:` lines, blank-line
|
||||
// separator) and dispatches Event values onto ch. Returns nil on
|
||||
// normal EOF.
|
||||
func parseSSE(ctx context.Context, r io.Reader, ch chan<- Event) error {
|
||||
scanner := bufio.NewScanner(r)
|
||||
scanner.Buffer(make([]byte, 64*1024), 1<<20) // 1 MiB line cap
|
||||
|
||||
var dataBuf bytes.Buffer
|
||||
for scanner.Scan() {
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return ctx.Err()
|
||||
default:
|
||||
}
|
||||
line := scanner.Bytes()
|
||||
if len(line) == 0 {
|
||||
// End-of-frame: dispatch buffered data, reset.
|
||||
if dataBuf.Len() == 0 {
|
||||
continue
|
||||
}
|
||||
var ev Event
|
||||
if err := json.Unmarshal(dataBuf.Bytes(), &ev); err == nil {
|
||||
select {
|
||||
case ch <- ev:
|
||||
case <-ctx.Done():
|
||||
return ctx.Err()
|
||||
}
|
||||
}
|
||||
dataBuf.Reset()
|
||||
continue
|
||||
}
|
||||
if bytes.HasPrefix(line, []byte("data:")) {
|
||||
payload := bytes.TrimSpace(line[5:])
|
||||
if dataBuf.Len() > 0 {
|
||||
dataBuf.WriteByte('\n')
|
||||
}
|
||||
dataBuf.Write(payload)
|
||||
}
|
||||
// `event:` lines describe the event name. JSON `data:` payload
|
||||
// also carries `type` field — we use that. Ignore event: lines.
|
||||
}
|
||||
if err := scanner.Err(); err != nil {
|
||||
return fmt.Errorf("sse scan: %w", err)
|
||||
}
|
||||
// Flush any trailing data without final blank line.
|
||||
if dataBuf.Len() > 0 {
|
||||
var ev Event
|
||||
if err := json.Unmarshal(dataBuf.Bytes(), &ev); err == nil {
|
||||
select {
|
||||
case ch <- ev:
|
||||
case <-ctx.Done():
|
||||
return ctx.Err()
|
||||
}
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
143
internal/anthropic/client_test.go
Normal file
143
internal/anthropic/client_test.go
Normal file
@@ -0,0 +1,143 @@
|
||||
package anthropic
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"context"
|
||||
"encoding/json"
|
||||
"io"
|
||||
"net/http"
|
||||
"net/http/httptest"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
func TestParseSSESingleEvent(t *testing.T) {
|
||||
body := "event: message_start\ndata: {\"type\":\"message_start\",\"message\":{\"id\":\"m1\",\"model\":\"X\"}}\n\n"
|
||||
ch := make(chan Event, 4)
|
||||
if err := parseSSE(context.Background(), strings.NewReader(body), ch); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
close(ch)
|
||||
events := drain(ch)
|
||||
if len(events) != 1 || events[0].Type != "message_start" || events[0].Message.ID != "m1" {
|
||||
t.Errorf("events = %+v", events)
|
||||
}
|
||||
}
|
||||
|
||||
func TestParseSSEMultiLineData(t *testing.T) {
|
||||
// "data:" can be split across multiple lines per spec; joined with '\n'.
|
||||
body := "data: {\"type\":\n" + "data: \"text_delta\",\"delta\":{\"type\":\"text_delta\",\"text\":\"hi\"}}\n\n"
|
||||
ch := make(chan Event, 4)
|
||||
if err := parseSSE(context.Background(), strings.NewReader(body), ch); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
close(ch)
|
||||
events := drain(ch)
|
||||
if len(events) != 1 || events[0].Type != "text_delta" || events[0].Delta.Text != "hi" {
|
||||
t.Errorf("events = %+v", events)
|
||||
}
|
||||
}
|
||||
|
||||
func TestParseSSEFlushTrailingWithoutBlank(t *testing.T) {
|
||||
body := "data: {\"type\":\"message_stop\"}"
|
||||
ch := make(chan Event, 4)
|
||||
if err := parseSSE(context.Background(), strings.NewReader(body), ch); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
close(ch)
|
||||
events := drain(ch)
|
||||
if len(events) != 1 || events[0].Type != "message_stop" {
|
||||
t.Errorf("trailing flush missed: %+v", events)
|
||||
}
|
||||
}
|
||||
|
||||
func TestStreamMessagesSendsRightHeaders(t *testing.T) {
|
||||
var gotAuth, gotVer string
|
||||
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
gotAuth = r.Header.Get("authorization")
|
||||
gotVer = r.Header.Get("anthropic-version")
|
||||
w.Header().Set("content-type", "text/event-stream")
|
||||
w.Write([]byte("event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n"))
|
||||
}))
|
||||
defer srv.Close()
|
||||
c := New(srv.URL, "fake-key")
|
||||
ch, err := c.StreamMessages(context.Background(), MessagesRequest{
|
||||
Model: "X", MaxTokens: 10,
|
||||
Messages: []Message{{Role: "user", Content: []ContentBlock{{"type": "text", "text": "hi"}}}},
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
for range ch {
|
||||
}
|
||||
if gotAuth != "Bearer fake-key" {
|
||||
t.Errorf("auth = %q", gotAuth)
|
||||
}
|
||||
if gotVer != DefaultAPIVersion {
|
||||
t.Errorf("anthropic-version = %q", gotVer)
|
||||
}
|
||||
}
|
||||
|
||||
func TestStreamMessagesNon2xxErrors(t *testing.T) {
|
||||
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
w.WriteHeader(http.StatusUnauthorized)
|
||||
w.Write([]byte(`{"error":{"type":"authentication_error","message":"bad key"}}`))
|
||||
}))
|
||||
defer srv.Close()
|
||||
c := New(srv.URL, "bad")
|
||||
_, err := c.StreamMessages(context.Background(), MessagesRequest{Model: "X", MaxTokens: 10})
|
||||
if err == nil || !strings.Contains(err.Error(), "401") {
|
||||
t.Errorf("err = %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestStreamMessagesStreamErrorBecomesEvent(t *testing.T) {
|
||||
// Simulate an SSE stream that errors mid-stream.
|
||||
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
w.Header().Set("content-type", "text/event-stream")
|
||||
flusher, _ := w.(http.Flusher)
|
||||
w.Write([]byte("event: message_start\ndata: {\"type\":\"message_start\"}\n\n"))
|
||||
flusher.Flush()
|
||||
// Hijack-style mid-stream close: not really an error in HTTP terms,
|
||||
// scanner just hits EOF cleanly. So no error event expected.
|
||||
}))
|
||||
defer srv.Close()
|
||||
c := New(srv.URL, "k")
|
||||
ch, err := c.StreamMessages(context.Background(), MessagesRequest{Model: "X", MaxTokens: 10})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
events := drain(ch)
|
||||
if len(events) != 1 || events[0].Type != "message_start" {
|
||||
t.Errorf("events = %+v", events)
|
||||
}
|
||||
}
|
||||
|
||||
func TestMessagesRequestMarshalShape(t *testing.T) {
|
||||
req := MessagesRequest{
|
||||
Model: "MiniMax-M2.7", MaxTokens: 50, Stream: true,
|
||||
Messages: []Message{{Role: "user", Content: []ContentBlock{
|
||||
{"type": "text", "text": "hi"},
|
||||
}}},
|
||||
}
|
||||
raw, _ := json.Marshal(req)
|
||||
// stream:true present
|
||||
if !bytes.Contains(raw, []byte(`"stream":true`)) {
|
||||
t.Errorf("missing stream:true in %s", raw)
|
||||
}
|
||||
if !bytes.Contains(raw, []byte(`"model":"MiniMax-M2.7"`)) {
|
||||
t.Errorf("missing model: %s", raw)
|
||||
}
|
||||
}
|
||||
|
||||
func drain(ch <-chan Event) []Event {
|
||||
var out []Event
|
||||
for ev := range ch {
|
||||
out = append(out, ev)
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
// Compile-time check that io.EOF is exported so we don't accidentally
|
||||
// remove the import.
|
||||
var _ = io.EOF
|
||||
317
internal/translate/translate.go
Normal file
317
internal/translate/translate.go
Normal file
@@ -0,0 +1,317 @@
|
||||
// Package translate converts between Plexum's canonical.TurnRequest /
|
||||
// canonical.TurnEvent shapes and the Anthropic Messages API shapes the
|
||||
// internal/anthropic client emits.
|
||||
//
|
||||
// Round-trip-able for text + thinking blocks; tool_use / tool_result
|
||||
// passes through structurally. Block-level signatures (the opaque
|
||||
// thinking signature Anthropic issues) are preserved when present.
|
||||
package translate
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
|
||||
"git.hangman-lab.top/hzhang/Plexum-sdk-go/canonical"
|
||||
|
||||
"git.hangman-lab.top/hzhang/Plexum-kimi-provider/internal/anthropic"
|
||||
)
|
||||
|
||||
// CanonicalToAnthropic converts a Plexum TurnRequest into an Anthropic
|
||||
// MessagesRequest. modelID overrides the canonical req.Model so callers
|
||||
// can map (e.g. "minimax/MiniMax-M2.7" → "MiniMax-M2.7") for the wire.
|
||||
func CanonicalToAnthropic(req canonical.TurnRequest, modelID string, defaultMaxTokens int) (anthropic.MessagesRequest, error) {
|
||||
out := anthropic.MessagesRequest{
|
||||
Model: modelID,
|
||||
MaxTokens: req.MaxTokens,
|
||||
}
|
||||
if out.MaxTokens <= 0 {
|
||||
out.MaxTokens = defaultMaxTokens
|
||||
}
|
||||
if req.Temperature != 0 {
|
||||
t := req.Temperature
|
||||
out.Temperature = &t
|
||||
}
|
||||
if len(req.StopSequences) > 0 {
|
||||
out.StopSequences = append([]string{}, req.StopSequences...)
|
||||
}
|
||||
// System: Plexum stores System as []Block; Anthropic accepts string
|
||||
// or []ContentBlock. We always use the []block form when we have any
|
||||
// non-trivial blocks so cache_control etc. pass through losslessly.
|
||||
if len(req.System) > 0 {
|
||||
sysBlocks := make([]anthropic.ContentBlock, 0, len(req.System))
|
||||
for _, b := range req.System {
|
||||
sysBlocks = append(sysBlocks, blockToAnthropic(b))
|
||||
}
|
||||
out.System = sysBlocks
|
||||
}
|
||||
// Messages.
|
||||
for _, m := range req.Messages {
|
||||
am := anthropic.Message{Role: roleToAnthropic(m.Role)}
|
||||
for _, b := range m.Content {
|
||||
am.Content = append(am.Content, blockToAnthropic(b))
|
||||
}
|
||||
out.Messages = append(out.Messages, am)
|
||||
}
|
||||
// Tools.
|
||||
for _, t := range req.Tools {
|
||||
out.Tools = append(out.Tools, anthropic.ToolDef{
|
||||
Name: t.Name, Description: t.Description, InputSchema: t.InputSchema,
|
||||
})
|
||||
}
|
||||
if req.ToolChoice != nil {
|
||||
out.ToolChoice = &anthropic.ToolChoice{Type: req.ToolChoice.Type, Name: req.ToolChoice.Name}
|
||||
}
|
||||
if req.Thinking != nil {
|
||||
mode := "disabled"
|
||||
if req.Thinking.Enabled {
|
||||
mode = "enabled"
|
||||
}
|
||||
out.Thinking = &anthropic.ThinkingConfig{
|
||||
Type: mode,
|
||||
BudgetTokens: req.Thinking.Budget,
|
||||
}
|
||||
}
|
||||
return out, nil
|
||||
}
|
||||
|
||||
func roleToAnthropic(r canonical.Role) string {
|
||||
switch r {
|
||||
case canonical.RoleUser:
|
||||
return "user"
|
||||
case canonical.RoleAssistant:
|
||||
return "assistant"
|
||||
default:
|
||||
return string(r)
|
||||
}
|
||||
}
|
||||
|
||||
// blockToAnthropic converts ONE canonical.Block into the loose
|
||||
// map-shaped ContentBlock Anthropic expects. Per-type handling:
|
||||
//
|
||||
// - TextBlock → {"type":"text", "text":..., "cache_control"?}
|
||||
// - ToolUseBlock → {"type":"tool_use", "id":..., "name":..., "input":...}
|
||||
// - ToolResultBlock → {"type":"tool_result", "tool_use_id":..., "content":[...], "is_error"?}
|
||||
// - ThinkingBlock → {"type":"thinking", "thinking":..., "signature":...}
|
||||
//
|
||||
// Unknown block types serialize to their JSON form via canonical's
|
||||
// own marshaller (fallback path).
|
||||
func blockToAnthropic(b canonical.Block) anthropic.ContentBlock {
|
||||
switch v := b.(type) {
|
||||
case *canonical.TextBlock:
|
||||
out := anthropic.ContentBlock{"type": "text", "text": v.Text}
|
||||
if v.CacheControl != nil {
|
||||
out["cache_control"] = v.CacheControl
|
||||
}
|
||||
return out
|
||||
case *canonical.ToolUseBlock:
|
||||
out := anthropic.ContentBlock{
|
||||
"type": "tool_use", "id": v.ID, "name": v.Name,
|
||||
}
|
||||
if len(v.Input) > 0 {
|
||||
out["input"] = json.RawMessage(v.Input)
|
||||
} else {
|
||||
out["input"] = map[string]any{}
|
||||
}
|
||||
return out
|
||||
case *canonical.ToolResultBlock:
|
||||
inner := make([]anthropic.ContentBlock, 0, len(v.Content))
|
||||
for _, ib := range v.Content {
|
||||
inner = append(inner, blockToAnthropic(ib))
|
||||
}
|
||||
out := anthropic.ContentBlock{
|
||||
"type": "tool_result", "tool_use_id": v.ToolUseID, "content": inner,
|
||||
}
|
||||
if v.IsError {
|
||||
out["is_error"] = true
|
||||
}
|
||||
return out
|
||||
case *canonical.ThinkingBlock:
|
||||
return anthropic.ContentBlock{
|
||||
"type": "thinking", "thinking": v.Thinking, "signature": v.Signature,
|
||||
}
|
||||
default:
|
||||
// Best-effort generic: marshal then unmarshal into a map.
|
||||
raw, err := json.Marshal(b)
|
||||
if err != nil {
|
||||
return anthropic.ContentBlock{"type": "text", "text": fmt.Sprintf("[unsupported block: %T]", b)}
|
||||
}
|
||||
var m map[string]any
|
||||
if err := json.Unmarshal(raw, &m); err != nil {
|
||||
return anthropic.ContentBlock{"type": "text", "text": string(raw)}
|
||||
}
|
||||
return m
|
||||
}
|
||||
}
|
||||
|
||||
// Translator turns a stream of anthropic.Event into a stream of
|
||||
// canonical.TurnEvent. Tracks per-content-block state (current type +
|
||||
// id + name) because Anthropic's deltas carry only the block index.
|
||||
type Translator struct {
|
||||
// Per-block tracking — index → metadata.
|
||||
blocks map[int]*blockState
|
||||
finalReason canonical.StopReason
|
||||
finalUsage canonical.Usage
|
||||
}
|
||||
|
||||
type blockState struct {
|
||||
Kind string // "text" | "thinking" | "tool_use"
|
||||
ID string // tool_use only
|
||||
Name string // tool_use only
|
||||
}
|
||||
|
||||
// NewTranslator constructs a fresh per-turn Translator.
|
||||
func NewTranslator() *Translator {
|
||||
return &Translator{blocks: map[int]*blockState{}}
|
||||
}
|
||||
|
||||
// Translate consumes one anthropic.Event and returns 0..N
|
||||
// canonical.TurnEvents. The translator owns no internal channel —
|
||||
// caller drives the loop.
|
||||
func (t *Translator) Translate(ev anthropic.Event) []canonical.TurnEvent {
|
||||
switch ev.Type {
|
||||
case "message_start":
|
||||
out := []canonical.TurnEvent{{Type: canonical.EventMessageStart}}
|
||||
if ev.Message != nil && ev.Message.Usage != nil {
|
||||
t.finalUsage = toCanonicalUsage(ev.Message.Usage)
|
||||
}
|
||||
return out
|
||||
|
||||
case "ping":
|
||||
return nil // keepalive
|
||||
|
||||
case "content_block_start":
|
||||
if ev.ContentBlock == nil {
|
||||
return nil
|
||||
}
|
||||
st := &blockState{Kind: ev.ContentBlock.Type, ID: ev.ContentBlock.ID, Name: ev.ContentBlock.Name}
|
||||
t.blocks[ev.Index] = st
|
||||
switch st.Kind {
|
||||
case "tool_use":
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventToolCallStart, ToolCallID: st.ID, ToolName: st.Name,
|
||||
}}
|
||||
case "text":
|
||||
// Some servers emit content_block_start with non-empty
|
||||
// text seed; surface that.
|
||||
if ev.ContentBlock.Text != "" {
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventTextDelta, Text: ev.ContentBlock.Text,
|
||||
}}
|
||||
}
|
||||
case "thinking":
|
||||
if ev.ContentBlock.Thinking != "" {
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventThinkingDelta, Thinking: ev.ContentBlock.Thinking,
|
||||
}}
|
||||
}
|
||||
}
|
||||
return nil
|
||||
|
||||
case "content_block_delta":
|
||||
if ev.Delta == nil {
|
||||
return nil
|
||||
}
|
||||
st := t.blocks[ev.Index]
|
||||
switch ev.Delta.Type {
|
||||
case "text_delta":
|
||||
return []canonical.TurnEvent{{Type: canonical.EventTextDelta, Text: ev.Delta.Text}}
|
||||
case "thinking_delta":
|
||||
return []canonical.TurnEvent{{Type: canonical.EventThinkingDelta, Thinking: ev.Delta.Thinking}}
|
||||
case "input_json_delta":
|
||||
if st == nil {
|
||||
return nil
|
||||
}
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventToolCallDelta,
|
||||
ToolCallID: st.ID, ToolName: st.Name,
|
||||
PartialJSON: ev.Delta.PartialJSON,
|
||||
}}
|
||||
case "signature_delta":
|
||||
// Signature lands on EventThinkingEnd at block_stop time;
|
||||
// stash on the block state.
|
||||
if st != nil {
|
||||
st.Name = ev.Delta.Signature // reuse Name as scratch
|
||||
}
|
||||
return nil
|
||||
}
|
||||
return nil
|
||||
|
||||
case "content_block_stop":
|
||||
st := t.blocks[ev.Index]
|
||||
if st == nil {
|
||||
return nil
|
||||
}
|
||||
delete(t.blocks, ev.Index)
|
||||
switch st.Kind {
|
||||
case "tool_use":
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventToolCallEnd, ToolCallID: st.ID, ToolName: st.Name,
|
||||
}}
|
||||
case "thinking":
|
||||
// st.Name was repurposed to carry the signature in
|
||||
// signature_delta above. Empty when no signature came.
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventThinkingEnd, Signature: st.Name,
|
||||
}}
|
||||
}
|
||||
return nil
|
||||
|
||||
case "message_delta":
|
||||
// Carries stop_reason + cumulative usage.
|
||||
if ev.Delta != nil && ev.Delta.StopReason != "" {
|
||||
t.finalReason = canonicalStopReason(ev.Delta.StopReason)
|
||||
}
|
||||
if ev.Usage != nil {
|
||||
// message_delta usage is INCREMENTAL on Anthropic; the v4 docs
|
||||
// describe it as cumulative across the stream. We just take
|
||||
// the last reported values verbatim.
|
||||
t.finalUsage = toCanonicalUsage(ev.Usage)
|
||||
}
|
||||
return nil
|
||||
|
||||
case "message_stop":
|
||||
usage := t.finalUsage
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventMessageEnd,
|
||||
StopReason: t.finalReason,
|
||||
Usage: &usage,
|
||||
}}
|
||||
|
||||
case "error":
|
||||
msg := "unknown"
|
||||
etype := "stream_error"
|
||||
if ev.Error != nil {
|
||||
msg = ev.Error.Message
|
||||
etype = ev.Error.Type
|
||||
}
|
||||
return []canonical.TurnEvent{{
|
||||
Type: canonical.EventError,
|
||||
Error: &canonical.ErrorInfo{Code: etype, Message: msg},
|
||||
}}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func canonicalStopReason(s string) canonical.StopReason {
|
||||
switch s {
|
||||
case "end_turn":
|
||||
return canonical.StopEndTurn
|
||||
case "max_tokens":
|
||||
return canonical.StopMaxTokens
|
||||
case "tool_use":
|
||||
return canonical.StopToolUse
|
||||
case "stop_sequence":
|
||||
return canonical.StopStopSequence
|
||||
default:
|
||||
return canonical.StopReason(s)
|
||||
}
|
||||
}
|
||||
|
||||
func toCanonicalUsage(u *anthropic.Usage) canonical.Usage {
|
||||
return canonical.Usage{
|
||||
InputTokens: u.InputTokens,
|
||||
OutputTokens: u.OutputTokens,
|
||||
CacheReadTokens: u.CacheReadInputTokens,
|
||||
CacheWriteTokens: u.CacheCreationInputTokens,
|
||||
}
|
||||
}
|
||||
77
scripts/install.sh
Executable file
77
scripts/install.sh
Executable file
@@ -0,0 +1,77 @@
|
||||
#!/usr/bin/env bash
|
||||
# Plexum-kimi-provider installer (Phase v0.1).
|
||||
#
|
||||
# Builds + installs:
|
||||
# ~/.plexum/plugins/plexum-kimi-provider/plexum-kimi-provider-plugin
|
||||
# ~/.plexum/plugins/plexum-kimi-provider/manifest.json
|
||||
#
|
||||
# Operator then writes the per-profile config:
|
||||
# ~/.plexum/plugins/plexum-kimi-provider/config.json
|
||||
# {"api_key": "sk-kimi-..."}
|
||||
#
|
||||
# Re-runnable. Profile data + config.json are never touched.
|
||||
#
|
||||
# Flags:
|
||||
# --profile <p> Override profile root (default ~/.plexum)
|
||||
set -euo pipefail
|
||||
|
||||
REPO="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/.." && pwd)"
|
||||
PROFILE_DIR="${HOME}/.plexum"
|
||||
|
||||
while [[ $# -gt 0 ]]; do
|
||||
case "$1" in
|
||||
--profile) PROFILE_DIR="$2"; shift 2 ;;
|
||||
-h|--help) sed -n '2,/^set -euo/p' "$0" | sed -n '/^#/p' | sed 's/^# \{0,1\}//'; exit 0 ;;
|
||||
*) echo "unknown flag: $1" >&2; exit 2 ;;
|
||||
esac
|
||||
done
|
||||
|
||||
log() { printf '\033[1;34m[kimi-install]\033[0m %s\n' "$*"; }
|
||||
command -v go >/dev/null || { echo "go not found on PATH" >&2; exit 1; }
|
||||
|
||||
PLUGIN_DIR="${PROFILE_DIR}/plugins/plexum-kimi-provider"
|
||||
mkdir -p "${PLUGIN_DIR}"
|
||||
|
||||
cd "${REPO}"
|
||||
VERSION="$(git describe --tags --always 2>/dev/null || echo dev)"
|
||||
LDFLAGS="-X main.Version=${VERSION}"
|
||||
log "building plexum-kimi-provider-plugin (v=${VERSION})"
|
||||
CGO_ENABLED=0 go build -ldflags="${LDFLAGS}" \
|
||||
-o "${PLUGIN_DIR}/plexum-kimi-provider-plugin" \
|
||||
./cmd/plexum-kimi-provider-plugin
|
||||
|
||||
cat > "${PLUGIN_DIR}/manifest.json" <<'EOF'
|
||||
{
|
||||
"name": "plexum-kimi-provider",
|
||||
"version": "0.1.0",
|
||||
"activation": "lazy",
|
||||
"executable": "plexum-kimi-provider-plugin",
|
||||
"contracts": {
|
||||
"provider": {
|
||||
"models": ["kimi-for-coding", "kimi-code", "k2p5"]
|
||||
}
|
||||
}
|
||||
}
|
||||
EOF
|
||||
|
||||
cat <<EOF
|
||||
|
||||
[kimi-install] done.
|
||||
plugin binary: ${PLUGIN_DIR}/plexum-kimi-provider-plugin
|
||||
manifest: ${PLUGIN_DIR}/manifest.json
|
||||
models: kimi-for-coding (+ legacy aliases: kimi-code, k2p5)
|
||||
endpoint: https://api.kimi.com/coding/v1/messages
|
||||
|
||||
Next steps:
|
||||
1. Write API key:
|
||||
cat > ${PLUGIN_DIR}/config.json <<JSON
|
||||
{"api_key": "sk-kimi-..."}
|
||||
JSON
|
||||
chmod 600 ${PLUGIN_DIR}/config.json
|
||||
2. Allow plugin:
|
||||
Add "plexum-kimi-provider" to ${PROFILE_DIR}/plexum.json's
|
||||
.plugins.allow array
|
||||
3. Point an agent at Kimi:
|
||||
plexum agent-add --model kimi-for-coding <agent-id>
|
||||
4. systemctl --user restart plexum
|
||||
EOF
|
||||
Reference in New Issue
Block a user