lam-agent-vicinae/src/lib/types.ts

/**
 * Shared domain types for Lam Agent.
 *
 * These are deliberately split into two families:
 *
 * - `ApiMessage` and friends mirror exactly what Ollama's `/api/chat` accepts.
 * - `Session`, `ChatTurn` and friends are Lam's own storage model.
 *
 * Keeping them separate matters because the storage model is derived from the
 * API model, not identical to it: the tool-calling messages that Ollama needs
 * mid-turn are not persisted, so a reloaded session replays cleanly as plain
 * user/assistant pairs.
 */

export type ChatRole = "system" | "user" | "assistant" | "tool";

export interface ToolCall {
  function: {
    name: string;
    arguments: Record<string, unknown>;
  };
}

export interface ToolDefinition {
  type: "function";
  function: {
    name: string;
    description: string;
    parameters: Record<string, unknown>;
  };
}

/** A message in the shape Ollama's `/api/chat` endpoint expects. */
export interface ApiMessage {
  role: ChatRole;
  content: string;
  tool_calls?: ToolCall[];
  /** Only set on `role: "tool"` messages, naming the tool that produced this content. */
  tool_name?: string;
  /**
   * Base64 image data, for vision models.
   *
   * Ollama only reads these on a `user` message (`gpt-oss` style tool roles have
   * no images). They are the one genuinely large thing in a request, so they are
   * attached to the question they belong to and never replayed later, the same
   * way tool output is not replayed.
   */
  images?: string[];
}

/**
 * A file the user attached to a question.
 *
 * Only the path is stored. A screenshot runs to megabytes, and
 * `MAX_STORED_CHARS_PER_TURN` exists precisely so one turn cannot dominate
 * LocalStorage, so the bytes stay on disk and are read when the turn is sent.
 */
export interface Attachment {
  /** Absolute path to the image on disk. */
  path: string;
  /** Sent to Ollama next to the base64 payload. */
  mime: string;
  /**
   * Pixel dimensions, read from the file's header.
   *
   * Kept because the transcript has to lay the preview out long after the file
   * was read, and Vicinae sizes an image's container to the request rather than
   * to the image, which leaves dead space unless the request has the right
   * aspect ratio. Absent for a format whose header we do not parse.
   */
  width?: number;
  height?: number;
}

/**
 * An attachment that has been read and is ready to send, held in the view while
 * the user types their question.
 *
 * The base64 payload lives here and nowhere else. It is needed for exactly one
 * request, so it is deliberately not part of `Attachment` and never reaches
 * storage.
 */
export interface PendingAttachment extends Attachment {
  /** Base64 of the image, with no `data:` URL prefix. Ollama wants it bare. */
  base64: string;
}
/** What a tool actually returned during a turn, kept for display and debugging. */
export interface ToolCallRecord {
  name: string;
  arguments: Record<string, unknown>;
  result: string;
  isError: boolean;
  /**
   * Set when the call was refused because the turn had already made it twice.
   *
   * Stored rather than inferred from the message text, so the transcript can say
   * what happened in its own words, and so a repeat that ended a turn stays
   * visible even when the user has hidden tool calls from the transcript.
   */
  repeated?: boolean;
}

/** One question and its answer. The unit the transcript is rendered from. */
export interface ChatTurn {
  id: string;
  prompt: string;
  response: string;
  thinking: string;
  model: string;
  createdAt: number;
  /** Set when the turn failed, so a stored session can show what went wrong. */
  error?: string;
  toolCalls: ToolCallRecord[];
  /** Images the user attached, kept for the transcript. The bytes are not stored. */
  attachments: Attachment[];
}

export interface Session {
  id: string;
  title: string;
  model: string;
  createdAt: number;
  updatedAt: number;
  turns: ChatTurn[];
}

/** The lightweight record kept in the session index, so listing sessions never loads full transcripts. */
export interface SessionSummary {
  id: string;
  title: string;
  model: string;
  createdAt: number;
  updatedAt: number;
  turnCount: number;
  preview: string;
  /**
   * Every image path any turn in this conversation refers to.
   *
   * Kept here rather than derived from the turns for one reason: the orphan
   * sweep needs to know which snips are still spoken for, and loading every
   * transcript to find out would mean a hundred JSON parses on every turn.
   *
   * Absent on entries written before this field existed, which is deliberately
   * not the same as an empty array: an absent value means "unknown", and the
   * sweep resolves it by reading that one transcript rather than by assuming the
   * conversation has no screenshots and deleting them.
   */
  attachmentPaths?: string[];
}

/**
 * Whether the model in use can read images, as far as the capability check knows.
 *
 * `"unknown"` is deliberately distinct from `"no"`: a failed request is not
 * evidence that a model cannot see, and reporting it as one would be a false
 * claim about the user's setup. Lives here rather than in `vision.ts` so that
 * modules which must stay free of Vicinae imports can name it.
 */
export type VisionSupport = "yes" | "no" | "unknown";

/**
 * The tool running right now, and how stale it is getting.
 *
 * Lives here rather than beside the chat state machine so that `thread.ts` can
 * carry it into the transcript, which is the only place it is now rendered.
 */
export interface ToolActivity {
  /** Human-readable label, e.g. "Searching the web". */
  label: string;
  /**
   * How many times this turn has now started this *identical* call.
   *
   * Counted by the agent over name and arguments together, so two different
   * searches are two tools doing their job rather than a repeat.
   */
  count: number;
}

/** A model as reported by `GET /api/tags`. */
export interface ModelInfo {
  name: string;
  parameterSize?: string;
  /**
   * How much disk the model takes, in bytes.
   *
   * Meaningless for a cloud model proxied through a local daemon: that entry is
   * a few hundred bytes of manifest, so the caller shows nothing rather than
   * "326 B", which is what `ollama list` does too.
   */
  size?: number;
  /**
   * What the model says it can do, e.g. `["tools", "vision"]`.
   *
   * Only newer servers send this. Absent means unknown, never "no": reporting a
   * missing field as a lack of vision would be a false claim about the user's
   * setup.
   */
  capabilities?: string[];
  /**
   * Set when this entry is a proxy rather than weights on disk, naming the server
   * that actually runs it (`https://ollama.com`). Deleting one of these drops a
   * reference; it does not free the gigabytes the size column implies.
   */
  remoteHost?: string;
}

/**
 * One rendered line of the conversation.
 *
 * A turn is stored as a question/answer pair, but the transcript is displayed as
 * a flat sequence of messages so each side can carry its own icon and weight.
 */
export interface ThreadMessage {
  id: string;
  role: "user" | "assistant";
  /** Full text, used for the detail panel and for copying. */
  text: string;
  createdAt: number;
  isStreaming: boolean;
  thinking: string;
  toolCalls: ToolCallRecord[];
  error?: string;
  /** Activity line shown while this message is streaming, e.g. "Searching the web". */
  status?: string;
  /**
   * The tool this message is waiting on, when there is one.
   *
   * Separate from {@link ThreadMessage.status} because the two are rendered from
   * different triggers: the status is a stand-in used only while there is no text
   * to read, while a running tool is worth saying whether or not tokens have
   * started arriving. Carrying both is what lets one line be printed for one
   * fact, rather than the same sentence appearing twice at the top of a turn.
   */
  activity?: ToolActivity;
  /** Images attached to this question. Only ever non-empty on a user message. */
  attachments: Attachment[];
}