Files
handler/frontend/lib/api.ts
T
Claude 2d5c0e34d7 Add local model backends: per-spawn dropdown pointing claude at alternative endpoints
Operators can register Anthropic-API-compatible endpoints (a local Qwen/Llama
behind LiteLLM or claude-code-router, an LLM gateway) on the dashboard's
Claude -> Models tab and pick one from a Model dropdown when spawning an agent.
The agent still launches as the same claude binary with the same hooks, skills,
connectors, plugins, and gates — only its ANTHROPIC_BASE_URL / ANTHROPIC_MODEL /
ANTHROPIC_AUTH_TOKEN env differs — and it stays pinned to its backend across
resumes. No selection keeps the worker's Claude subscription untouched.

- claude_models table (+ agents.model_id pin), migration 0012
- control.models resolves a row into the launch env (API keys Fernet-encrypted
  at rest, decrypted only in the control container; placeholder key when none is
  stored so the subscription OAuth token never reaches a local endpoint)
- /claude/models CRUD (admin-gated writes, key never returned), spawn route +
  worker + CLI (--model) pass the selection through, fail-fast on missing or
  disabled backends
- dashboard: Models tab, spawn-form dropdown, model badge in the agents table
- docs/local-models.md: why bare OpenAI-compatible servers break tool calling
  with Qwen-Coder, and working vLLM/LiteLLM/llama.cpp stacks

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DzDofD7gP63WpeLG8vEdZu
2026-07-29 18:36:16 +00:00

297 lines
8.4 KiB
TypeScript

/* Typed client for the Handler API + the row shapes it returns (mirrors the FastAPI
* pydantic schemas in src/handler/api/schemas.py). The browser calls the API same-origin
* with relative paths; set NEXT_PUBLIC_API_BASE to point `npm run dev` at another origin.
*
* NOTE: this file lives under frontend/lib/, now un-ignored in .gitignore so the source
* ships and the build works from a fresh clone (the built export under
* src/handler/api/static/ is what the package serves). */
const BASE = process.env.NEXT_PUBLIC_API_BASE ?? "";
export type CommandStatus = "queued" | "running" | "done" | "failed";
export interface Project {
id: string;
root_dir: string;
git_remote?: string | null;
credential_ref?: string | null;
created_at: string;
/* Present on the registration response in git-server mode: the enqueued clone. */
sync_command_id?: number | null;
/* Present on the registration response when "Initialize mise" was ticked: the
* enqueued bootstrap agent that writes + commits + pushes a .mise.toml. */
mise_init_command_id?: number | null;
}
export interface Agent {
id: number;
project_id: string;
name: string;
working_dir: string;
status: string;
role?: string | null;
/* Model backend the agent is pinned to (see ClaudeModel); null = the Claude
* subscription the worker is logged in to. */
model_id?: number | null;
/* Latest output snapshot from the worker: the tmux pane tail for legacy agents, the
* latest assistant text for headless runs. For a crashed agent this is the evidence
* frame — the last thing the process said. */
last_output?: string | null;
output_at?: string | null;
/* Headless runner: claude session UUID (null = legacy tmux agent) + supervising worker. */
session_id?: string | null;
worker_id?: string | null;
created_at: string;
}
/* One persisted stream-json event of a headless run (GET .../events, cursor-paged by id).
* `type` mirrors the stream (system/assistant/user/result) plus `worker` (runner notices)
* and `raw` (unparseable line kept verbatim). */
export interface AgentEvent {
id: number;
agent_id: number;
run_id: number;
session_id?: string | null;
seq: number;
type: string;
payload?: Record<string, unknown> | null;
created_at: string;
}
export interface Checkmark {
agent_id: number;
checkpoint_at: string;
status: string;
where_it_stopped?: string | null;
next_steps?: string[] | null;
open_question?: string | null;
log_entry_id?: number | null;
tests_status: string;
tested_at?: string | null;
build_status: string;
built_at?: string | null;
}
export interface LogEntry {
id: number;
agent_id: number;
created_at: string;
session_id?: string | null;
status: string;
summary?: string | null;
decisions?: string | null;
question?: string | null;
answer?: string | null;
visibility: string;
push_sha?: string | null;
ci_status: string;
ci_checked_at?: string | null;
}
export interface Approval {
id: number;
project_id: string;
branch: string;
approved_sha?: string | null;
pr_ref?: string | null;
status: string;
approved_by_agent_id?: number | null;
actor?: string | null;
note?: string | null;
created_at: string;
}
export interface Host {
hostname: string;
forge_type: string;
token_env_var?: string | null;
base_url?: string | null;
ssh_public_key?: string | null;
has_token: boolean;
created_at: string;
}
export interface Command {
id: number;
project_id?: string | null;
agent_name?: string | null;
type: string;
payload?: Record<string, unknown> | null;
status: CommandStatus;
result?: Record<string, unknown> | null;
error?: string | null;
requested_by?: string | null;
claimed_by?: string | null;
created_at: string;
claimed_at?: string | null;
finished_at?: string | null;
}
export interface Schedule {
id: number;
project_id: string;
name_prefix: string;
task: string;
role?: string | null;
worktree?: string | null;
subdir?: string | null;
interval_seconds: number;
enabled: boolean;
next_run_at: string;
last_run_at?: string | null;
last_command_id?: number | null;
created_at: string;
}
/* ---- Claude management (the dashboard's Claude page) ---- */
export interface ClaudeSkill {
id: number;
name: string;
description?: string | null;
content: string;
enabled: boolean;
/* Relative paths of auxiliary files captured by an install-from-prompt import
* (references/, scripts/, …); synced alongside SKILL.md, read-only here. */
files: string[];
created_at: string;
updated_at: string;
}
export type McpTransport = "stdio" | "http" | "sse";
export interface ClaudeConnector {
id: number;
name: string;
transport: McpTransport;
command?: string | null;
args?: string[] | null;
env?: Record<string, string> | null;
url?: string | null;
headers?: Record<string, string> | null;
enabled: boolean;
created_at: string;
}
export interface ClaudePlugin {
id: number;
name: string;
marketplace: string;
marketplace_repo: string;
enabled: boolean;
created_at: string;
}
/* A registered model backend: an Anthropic-API-compatible endpoint (a local model
* behind LiteLLM / claude-code-router, an LLM gateway) the spawn dropdown offers next
* to the Claude subscription. The API key is write-only server-side (has_api_key only). */
export interface ClaudeModel {
id: number;
name: string;
base_url: string;
model: string;
small_fast_model?: string | null;
env?: Record<string, string> | null;
enabled: boolean;
has_api_key: boolean;
created_at: string;
}
/* Stored overrides + the env baseline they merge over at launch (read-only here). */
export interface ClaudePermissions {
default_mode?: string | null;
allow: string[];
deny: string[];
ask: string[];
base_mode: string;
base_allow: string[];
}
export interface SharedContext {
key: string;
value: string;
set_by_agent_id?: number | null;
updated_at: string;
}
/* Thrown on a 401 so callers can distinguish "token rejected" from real errors and stay
* quiet while the app re-prompts for a token. */
export class AuthError extends Error {
constructor(message = "unauthorized") {
super(message);
this.name = "AuthError";
}
}
/* Any non-2xx (other than 401); carries the HTTP status so callers can branch on 404 etc. */
export interface ApiError extends Error {
status: number;
}
interface ApiOptions {
method?: string;
body?: unknown;
}
interface TrackOptions {
attempts?: number;
intervalMs?: number;
}
export interface ApiClient {
api: <T>(path: string, opts?: ApiOptions) => Promise<T>;
/* Poll GET /commands/{id} until it reaches done/failed; null if still running after the
* budget (worker down or a very slow command). */
trackCommand: (id: number, opts?: TrackOptions) => Promise<Command | null>;
}
export function createClient(token: string, onUnauthorized: () => void): ApiClient {
async function api<T>(path: string, opts?: ApiOptions): Promise<T> {
const hasBody = opts?.body !== undefined && opts?.body !== null;
const res = await fetch(BASE + path, {
method: opts?.method ?? (hasBody ? "POST" : "GET"),
headers: {
Authorization: `Bearer ${token}`,
...(hasBody ? { "Content-Type": "application/json" } : {}),
},
body: hasBody ? JSON.stringify(opts!.body) : undefined,
});
if (res.status === 401) {
onUnauthorized();
throw new AuthError();
}
if (!res.ok) {
let detail: string = res.statusText;
try {
const j = await res.json();
if (j && typeof j.detail !== "undefined") {
detail = typeof j.detail === "string" ? j.detail : JSON.stringify(j.detail);
}
} catch {
/* non-JSON error body; keep statusText */
}
const err = new Error(detail) as ApiError;
err.status = res.status;
throw err;
}
if (res.status === 204) return undefined as T;
const text = await res.text();
return (text ? JSON.parse(text) : undefined) as T;
}
async function trackCommand(id: number, opts?: TrackOptions): Promise<Command | null> {
const attempts = opts?.attempts ?? 60;
const intervalMs = opts?.intervalMs ?? 500;
for (let i = 0; i < attempts; i++) {
const cmd = await api<Command>(`/commands/${id}`);
if (cmd.status === "done" || cmd.status === "failed") return cmd;
await new Promise((r) => setTimeout(r, intervalMs));
}
return null;
}
return { api, trackCommand };
}