@sunilp-org/jam-cli 0.1.0 → 0.1.2

package/dist/utils/memory.js

@@ -0,0 +1,215 @@
+/**
+ * Working memory management for the agentic loop.
+ *
+ * Provides:
+ * - **Context compaction** — summarize old tool results when approaching the
+ *   context window limit.
+ * - **Scratchpad** — periodic "what have I learned" prompts that create a
+ *   running working memory the model can reference.
+ * - **Tool result capping** — truncate oversized tool outputs before injection.
+ *
+ * The core idea: instead of feeding the LLM an ever-growing message array,
+ * we periodically compress old rounds into a compact summary and carry only
+ * recent rounds + the summary forward.
+ */
+import { checkBudget, truncateToolOutput, } from './tokens.js';
+// ── Configuration ─────────────────────────────────────────────────────────────
+/** After this many tool rounds, inject a scratchpad prompt. */
+export const SCRATCHPAD_INTERVAL = 3;
+/** When messages exceed this fraction of the context budget, compact. */
+const COMPACTION_THRESHOLD = 0.70;
+/** Max token budget for a single tool result injection. */
+export const MAX_TOOL_RESULT_TOKENS = 1500;
+/** Max token budget for compacted summary. */
+const _MAX_SUMMARY_TOKENS = 800;
+// ── Tool result capping (P0-2) ────────────────────────────────────────────────
+/**
+ * Cap a tool output to prevent oversized injections into the message array.
+ * Call this BEFORE pushing the tool result into `messages`.
+ */
+export function capToolResult(toolName, output) {
+    return truncateToolOutput(toolName, output, MAX_TOOL_RESULT_TOKENS);
+}
+// ── Scratchpad (P1-4) ────────────────────────────────────────────────────────
+const SCRATCHPAD_PROMPT = `[WORKING MEMORY CHECKPOINT]
+
+Pause and organize what you have learned so far.  Write a brief, structured note:
+
+1. **Files examined**: List every file path you have read or searched.
+2. **Key findings**: List the most important facts you found (function names, patterns, locations).
+3. **Still needed**: What information do you still need to answer the user's question?
+
+Keep this under 200 words.  This note will stay in your context as working memory.`;
+/**
+ * Check if it's time to inject a scratchpad prompt.
+ * Returns true every SCRATCHPAD_INTERVAL tool rounds.
+ */
+export function shouldInjectScratchpad(toolRound) {
+    return toolRound > 0 && toolRound % SCRATCHPAD_INTERVAL === 0;
+}
+/**
+ * Build the scratchpad injection message.
+ */
+export function buildScratchpadPrompt() {
+    return { role: 'user', content: SCRATCHPAD_PROMPT };
+}
+// ── Context compaction / summarization (P0-3) ─────────────────────────────────
+/**
+ * Check whether the message array needs compaction.
+ */
+export function needsCompaction(messages, systemPrompt, model) {
+    const { budget, currentTokens } = checkBudget(messages, systemPrompt, model);
+    return currentTokens > budget * COMPACTION_THRESHOLD;
+}
+const SUMMARIZER_SYSTEM_PROMPT = `You are a context summarizer for a code assistant. You will receive a conversation between a user and an assistant that includes tool calls and their results.
+
+Your job is to produce a COMPACT summary of all the information gathered so far. Include:
+1. Files that were examined (paths only)
+2. Key code facts discovered (function names, class names, patterns, important line numbers)
+3. Any errors or dead-ends encountered
+
+Rules:
+- Be extremely concise — bullet points only
+- Include file paths and line numbers where relevant
+- Do NOT include opinions or analysis — just facts
+- Do NOT include the full contents of files — just what was found
+- Stay under 300 words`;
+/**
+ * Compact the message array by summarizing old tool rounds.
+ *
+ * Strategy:
+ * 1. Keep the first message (original user query) and the last N messages (recent context).
+ * 2. Summarize everything in between via a cheap LLM call.
+ * 3. Replace the middle messages with a single summary message.
+ *
+ * Falls back to a simple truncation if the LLM call fails.
+ */
+export async function compactMessages(messages, provider, options) {
+    const keepRecent = options.keepRecent ?? 6; // Keep last 6 messages (3 rounds)
+    // Not enough messages to compact
+    if (messages.length <= keepRecent + 2)
+        return messages;
+    const firstMessage = messages[0];
+    const middleMessages = messages.slice(1, -keepRecent);
+    const recentMessages = messages.slice(-keepRecent);
+    // Build a condensed representation of middle messages for the summarizer
+    const middleText = middleMessages
+        .map((m) => `[${m.role}] ${m.content.slice(0, 500)}${m.content.length > 500 ? '…' : ''}`)
+        .join('\n---\n');
+    // Try to summarize via LLM
+    try {
+        const summaryRequest = {
+            messages: [{
+                    role: 'user',
+                    content: `Summarize the following conversation context:\n\n${middleText}`,
+                }],
+            model: options.model,
+            temperature: 0.1, // Very focused
+            maxTokens: 500,
+            systemPrompt: SUMMARIZER_SYSTEM_PROMPT,
+        };
+        let summary = '';
+        const stream = provider.streamCompletion(summaryRequest);
+        for await (const chunk of stream) {
+            if (!chunk.done)
+                summary += chunk.delta;
+        }
+        const trimmedSummary = summary.trim();
+        if (trimmedSummary.length > 20) {
+            // Successfully summarized — replace middle with summary
+            const summaryMessage = {
+                role: 'user',
+                content: `[CONTEXT SUMMARY — earlier tool results compressed]\n\n${trimmedSummary}\n\n[End of summary — recent context follows]`,
+            };
+            return [firstMessage, summaryMessage, ...recentMessages];
+        }
+    }
+    catch {
+        // Summarization failed — fall back to simple truncation
+    }
+    // Fallback: just keep first + truncated middle + recent
+    const fallbackSummary = {
+        role: 'user',
+        content: `[CONTEXT NOTE: ${middleMessages.length} earlier messages were compressed to save context space. Key info may need to be re-discovered if not in recent messages.]`,
+    };
+    return [firstMessage, fallbackSummary, ...recentMessages];
+}
+// ── Combined memory manager ───────────────────────────────────────────────────
+/**
+ * WorkingMemory manages the conversation's context budget throughout
+ * the agentic loop.  Commands use it like:
+ *
+ * ```ts
+ * const memory = new WorkingMemory(provider, model, systemPrompt);
+ * // In the tool loop:
+ * memory.addToolResult(toolName, output, messages);
+ * if (memory.shouldScratchpad(round)) messages.push(memory.scratchpadPrompt());
+ * if (memory.shouldCompact(messages)) messages = await memory.compact(messages);
+ * ```
+ */
+export class WorkingMemory {
+    provider;
+    model;
+    systemPrompt;
+    readFiles = new Set();
+    searchQueries = new Set();
+    factsLearned = [];
+    constructor(provider, model, systemPrompt) {
+        this.provider = provider;
+        this.model = model;
+        this.systemPrompt = systemPrompt;
+    }
+    /**
+     * Process and cap a tool result before injection into messages.
+     * Also tracks which files/searches have been done.
+     */
+    processToolResult(toolName, args, output) {
+        // Track what we've accessed
+        if (toolName === 'read_file' && args['path']) {
+            this.readFiles.add(String(args['path']));
+        }
+        if (toolName === 'search_text' && args['query']) {
+            this.searchQueries.add(String(args['query']));
+        }
+        // Cap the output
+        return capToolResult(toolName, output);
+    }
+    /**
+     * Check if we should inject a scratchpad prompt at this round.
+     */
+    shouldScratchpad(round) {
+        return shouldInjectScratchpad(round);
+    }
+    /**
+     * Build the scratchpad prompt.
+     */
+    scratchpadPrompt() {
+        return buildScratchpadPrompt();
+    }
+    /**
+     * Check if messages need compaction.
+     */
+    shouldCompact(messages) {
+        return needsCompaction(messages, this.systemPrompt, this.model);
+    }
+    /**
+     * Compact the messages array.
+     */
+    async compact(messages) {
+        return compactMessages(messages, this.provider, { model: this.model });
+    }
+    /**
+     * Get a summary of what has been accessed (for diagnostics / JAM.md updates).
+     */
+    getAccessLog() {
+        return {
+            readFiles: [...this.readFiles],
+            searchQueries: [...this.searchQueries],
+        };
+    }
+    /** Reset for a new turn (multi-turn chat). */
+    reset() {
+        // Don't reset readFiles/searchQueries — they're session-level
+    }
+}
+//# sourceMappingURL=memory.js.map

package/dist/utils/memory.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"memory.js","sourceRoot":"","sources":["../../src/utils/memory.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAIH,OAAO,EACL,WAAW,EACX,kBAAkB,GACnB,MAAM,aAAa,CAAC;AAErB,iFAAiF;AAEjF,+DAA+D;AAC/D,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAErC,yEAAyE;AACzE,MAAM,oBAAoB,GAAG,IAAI,CAAC;AAElC,2DAA2D;AAC3D,MAAM,CAAC,MAAM,sBAAsB,GAAG,IAAI,CAAC;AAE3C,8CAA8C;AAC9C,MAAM,mBAAmB,GAAG,GAAG,CAAC;AAEhC,iFAAiF;AAEjF;;;GAGG;AACH,MAAM,UAAU,aAAa,CAAC,QAAgB,EAAE,MAAc;IAC5D,OAAO,kBAAkB,CAAC,QAAQ,EAAE,MAAM,EAAE,sBAAsB,CAAC,CAAC;AACtE,CAAC;AAED,gFAAgF;AAEhF,MAAM,iBAAiB,GAAG;;;;;;;;mFAQyD,CAAC;AAEpF;;;GAGG;AACH,MAAM,UAAU,sBAAsB,CAAC,SAAiB;IACtD,OAAO,SAAS,GAAG,CAAC,IAAI,SAAS,GAAG,mBAAmB,KAAK,CAAC,CAAC;AAChE,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB;IACnC,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,EAAE,CAAC;AACtD,CAAC;AAED,iFAAiF;AAEjF;;GAEG;AACH,MAAM,UAAU,eAAe,CAC7B,QAAmB,EACnB,YAAgC,EAChC,KAAc;IAEd,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,GAAG,WAAW,CAAC,QAAQ,EAAE,YAAY,EAAE,KAAK,CAAC,CAAC;IAC7E,OAAO,aAAa,GAAG,MAAM,GAAG,oBAAoB,CAAC;AACvD,CAAC;AAED,MAAM,wBAAwB,GAAG;;;;;;;;;;;;uBAYV,CAAC;AAExB;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,QAAmB,EACnB,QAAyB,EACzB,OAAgD;IAEhD,MAAM,UAAU,GAAG,OAAO,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC,kCAAkC;IAE9E,iCAAiC;IACjC,IAAI,QAAQ,CAAC,MAAM,IAAI,UAAU,GAAG,CAAC;QAAE,OAAO,QAAQ,CAAC;IAEvD,MAAM,YAAY,GAAG,QAAQ,CAAC,CAAC,CAAE,CAAC;IAClC,MAAM,cAAc,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,UAAU,CAAC,CAAC;IACtD,MAAM,cAAc,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,CAAC;IAEnD,yEAAyE;IACzE,MAAM,UAAU,GAAG,cAAc;SAC9B,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;SACxF,IAAI,CAAC,SAAS,CAAC,CAAC;IAEnB,2BAA2B;IAC3B,IAAI,CAAC;QACH,MAAM,cAAc,GAAG;YACrB,QAAQ,EAAE,CAAC;oBACT,IAAI,EAAE,MAAe;oBACrB,OAAO,EAAE,oDAAoD,UAAU,EAAE;iBAC1E,CAAC;YACF,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,WAAW,EAAE,GAAG,EAAE,eAAe;YACjC,SAAS,EAAE,GAAG;YACd,YAAY,EAAE,wBAAwB;SACvC,CAAC;QAEF,IAAI,OAAO,GAAG,EAAE,CAAC;QACjB,MAAM,MAAM,GAAG,QAAQ,CAAC,gBAAgB,CAAC,cAAc,CAAC,CAAC;QACzD,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YACjC,IAAI,CAAC,KAAK,CAAC,IAAI;gBAAE,OAAO,IAAI,KAAK,CAAC,KAAK,CAAC;QAC1C,CAAC;QAED,MAAM,cAAc,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC;QACtC,IAAI,cAAc,CAAC,MAAM,GAAG,EAAE,EAAE,CAAC;YAC/B,wDAAwD;YACxD,MAAM,cAAc,GAAY;gBAC9B,IAAI,EAAE,MAAM;gBACZ,OAAO,EAAE,0DAA0D,cAAc,+CAA+C;aACjI,CAAC;YAEF,OAAO,CAAC,YAAY,EAAE,cAAc,EAAE,GAAG,cAAc,CAAC,CAAC;QAC3D,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,wDAAwD;IAC1D,CAAC;IAED,wDAAwD;IACxD,MAAM,eAAe,GAAY;QAC/B,IAAI,EAAE,MAAM;QACZ,OAAO,EAAE,kBAAkB,cAAc,CAAC,MAAM,4HAA4H;KAC7K,CAAC;IAEF,OAAO,CAAC,YAAY,EAAE,eAAe,EAAE,GAAG,cAAc,CAAC,CAAC;AAC5D,CAAC;AAED,iFAAiF;AAEjF;;;;;;;;;;;GAWG;AACH,MAAM,OAAO,aAAa;IAChB,QAAQ,CAAkB;IAC1B,KAAK,CAAU;IACf,YAAY,CAAU;IACtB,SAAS,GAAgB,IAAI,GAAG,EAAE,CAAC;IACnC,aAAa,GAAgB,IAAI,GAAG,EAAE,CAAC;IACvC,YAAY,GAAa,EAAE,CAAC;IAEpC,YAAY,QAAyB,EAAE,KAAc,EAAE,YAAqB;QAC1E,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;IACnC,CAAC;IAED;;;OAGG;IACH,iBAAiB,CAAC,QAAgB,EAAE,IAA6B,EAAE,MAAc;QAC/E,4BAA4B;QAC5B,IAAI,QAAQ,KAAK,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;YAC7C,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC3C,CAAC;QACD,IAAI,QAAQ,KAAK,aAAa,IAAI,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;YAChD,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAChD,CAAC;QAED,iBAAiB;QACjB,OAAO,aAAa,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IACzC,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,KAAa;QAC5B,OAAO,sBAAsB,CAAC,KAAK,CAAC,CAAC;IACvC,CAAC;IAED;;OAEG;IACH,gBAAgB;QACd,OAAO,qBAAqB,EAAE,CAAC;IACjC,CAAC;IAED;;OAEG;IACH,aAAa,CAAC,QAAmB;QAC/B,OAAO,eAAe,CAAC,QAAQ,EAAE,IAAI,CAAC,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;IAClE,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,OAAO,CAAC,QAAmB;QAC/B,OAAO,eAAe,CAAC,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IACzE,CAAC;IAED;;OAEG;IACH,YAAY;QACV,OAAO;YACL,SAAS,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC;YAC9B,aAAa,EAAE,CAAC,GAAG,IAAI,CAAC,aAAa,CAAC;SACvC,CAAC;IACJ,CAAC;IAED,8CAA8C;IAC9C,KAAK;QACH,8DAA8D;IAChE,CAAC;CACF"}

package/dist/utils/past-sessions.d.ts

@@ -0,0 +1,31 @@
+/**
+ * Past session search — find relevant Q&A from previous sessions.
+ *
+ * Uses a simple TF-IDF-like keyword overlap to find past conversations
+ * that are relevant to the current question.  No vector embeddings needed —
+ * this is a pragmatic approach for local-first CLI tools.
+ */
+export interface PastExchange {
+    /** The user's question from the past session. */
+    question: string;
+    /** The assistant's answer from the past session. */
+    answer: string;
+    /** The session name / id. */
+    sessionId: string;
+    /** Relevance score (0–1). */
+    score: number;
+}
+/**
+ * Search past sessions for Q&A exchanges relevant to the current question.
+ *
+ * @param question          The user's current question.
+ * @param workspaceRoot     Current workspace root (to scope sessions).
+ * @param maxResults        Maximum exchanges to return.
+ * @param minScore          Minimum relevance score to include.
+ */
+export declare function searchPastSessions(question: string, workspaceRoot: string, maxResults?: number, minScore?: number): Promise<PastExchange[]>;
+/**
+ * Format past exchanges as context to inject into the system prompt or messages.
+ */
+export declare function formatPastExchanges(exchanges: PastExchange[]): string;
+//# sourceMappingURL=past-sessions.d.ts.map

package/dist/utils/past-sessions.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"past-sessions.d.ts","sourceRoot":"","sources":["../../src/utils/past-sessions.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAOH,MAAM,WAAW,YAAY;IAC3B,iDAAiD;IACjD,QAAQ,EAAE,MAAM,CAAC;IACjB,oDAAoD;IACpD,MAAM,EAAE,MAAM,CAAC;IACf,6BAA6B;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,6BAA6B;IAC7B,KAAK,EAAE,MAAM,CAAC;CACf;AAyCD;;;;;;;GAOG;AACH,wBAAsB,kBAAkB,CACtC,QAAQ,EAAE,MAAM,EAChB,aAAa,EAAE,MAAM,EACrB,UAAU,GAAE,MAAU,EACtB,QAAQ,GAAE,MAAa,GACtB,OAAO,CAAC,YAAY,EAAE,CAAC,CAwDzB;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,YAAY,EAAE,GAAG,MAAM,CAqBrE"}

package/dist/utils/past-sessions.js

@@ -0,0 +1,126 @@
+/**
+ * Past session search — find relevant Q&A from previous sessions.
+ *
+ * Uses a simple TF-IDF-like keyword overlap to find past conversations
+ * that are relevant to the current question.  No vector embeddings needed —
+ * this is a pragmatic approach for local-first CLI tools.
+ */
+import { listSessions, getSession } from '../storage/history.js';
+// ── Tokenization / scoring ────────────────────────────────────────────────────
+const STOP_WORDS = new Set([
+    'the', 'a', 'an', 'is', 'are', 'was', 'were', 'be', 'been', 'being',
+    'have', 'has', 'had', 'do', 'does', 'did', 'will', 'would', 'could',
+    'should', 'may', 'might', 'can', 'shall', 'to', 'of', 'in', 'for',
+    'on', 'with', 'at', 'by', 'from', 'as', 'into', 'through', 'about',
+    'this', 'that', 'and', 'or', 'but', 'not', 'so', 'if', 'then',
+    'than', 'too', 'very', 'just', 'how', 'where', 'what', 'which',
+    'who', 'when', 'why', 'all', 'each', 'every', 'some', 'any',
+    'its', 'it', 'you', 'your', 'we', 'our', 'they', 'their',
+    'i', 'me', 'my', 'change', 'adding', 'using', 'make', 'use',
+]);
+function tokenize(text) {
+    return text
+        .toLowerCase()
+        .replace(/[^a-z0-9\s_.-]/g, ' ')
+        .split(/\s+/)
+        .filter(w => w.length > 2 && !STOP_WORDS.has(w));
+}
+/**
+ * Compute a simple keyword overlap score between two texts.
+ * Returns 0–1 where 1 means perfect overlap.
+ */
+function keywordOverlap(queryTokens, targetTokens) {
+    if (queryTokens.length === 0 || targetTokens.length === 0)
+        return 0;
+    const targetSet = new Set(targetTokens);
+    const matches = queryTokens.filter(t => targetSet.has(t));
+    // Jaccard-like: overlap / union
+    const union = new Set([...queryTokens, ...targetTokens]);
+    return matches.length / union.size;
+}
+// ── Main search ───────────────────────────────────────────────────────────────
+/**
+ * Search past sessions for Q&A exchanges relevant to the current question.
+ *
+ * @param question          The user's current question.
+ * @param workspaceRoot     Current workspace root (to scope sessions).
+ * @param maxResults        Maximum exchanges to return.
+ * @param minScore          Minimum relevance score to include.
+ */
+export async function searchPastSessions(question, workspaceRoot, maxResults = 3, minScore = 0.15) {
+    const queryTokens = tokenize(question);
+    if (queryTokens.length === 0)
+        return [];
+    let sessions;
+    try {
+        sessions = await listSessions();
+    }
+    catch {
+        return [];
+    }
+    // Only look at sessions from the same workspace, limit to recent 20
+    const relevantSessions = sessions
+        .filter(s => s.workspaceRoot === workspaceRoot)
+        .slice(0, 20);
+    const candidates = [];
+    for (const sessionMeta of relevantSessions) {
+        let session;
+        try {
+            session = await getSession(sessionMeta.id);
+        }
+        catch {
+            continue;
+        }
+        if (!session?.messages)
+            continue;
+        // Extract user→assistant pairs
+        for (let i = 0; i < session.messages.length - 1; i++) {
+            const msg = session.messages[i];
+            const next = session.messages[i + 1];
+            if (msg.role === 'user' && next?.role === 'assistant') {
+                // Skip tool-result injections and system messages
+                if (msg.content.startsWith('[Tool result:') || msg.content.startsWith('[SYSTEM'))
+                    continue;
+                if (msg.content.startsWith('[WORKING MEMORY') || msg.content.startsWith('[CONTEXT'))
+                    continue;
+                const questionTokens = tokenize(msg.content);
+                const score = keywordOverlap(queryTokens, questionTokens);
+                if (score >= minScore) {
+                    candidates.push({
+                        question: msg.content.slice(0, 500),
+                        answer: next.content.slice(0, 1500), // Cap size
+                        sessionId: session.id,
+                        score,
+                    });
+                }
+            }
+        }
+    }
+    // Sort by score descending, take top N
+    return candidates
+        .sort((a, b) => b.score - a.score)
+        .slice(0, maxResults);
+}
+/**
+ * Format past exchanges as context to inject into the system prompt or messages.
+ */
+export function formatPastExchanges(exchanges) {
+    if (exchanges.length === 0)
+        return '';
+    const parts = [
+        '## Relevant Past Conversations',
+        '',
+        'These previous Q&A exchanges from this project may provide useful context:',
+        '',
+    ];
+    for (const ex of exchanges) {
+        parts.push(`**Q:** ${ex.question.slice(0, 200)}`);
+        parts.push(`**A:** ${ex.answer.slice(0, 500)}`);
+        parts.push('');
+    }
+    parts.push('---');
+    parts.push('Use the above as background context, but always verify by reading current code.');
+    parts.push('');
+    return parts.join('\n');
+}
+//# sourceMappingURL=past-sessions.js.map

package/dist/utils/past-sessions.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"past-sessions.js","sourceRoot":"","sources":["../../src/utils/past-sessions.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAgBjE,iFAAiF;AAEjF,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC;IACzB,KAAK,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO;IACnE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO;IACnE,QAAQ,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK;IACjE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO;IAClE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI,EAAE,MAAM;IAC7D,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO;IAC9D,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,KAAK;IAC3D,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO;IACxD,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,KAAK;CAC5D,CAAC,CAAC;AAEH,SAAS,QAAQ,CAAC,IAAY;IAC5B,OAAO,IAAI;SACR,WAAW,EAAE;SACb,OAAO,CAAC,iBAAiB,EAAE,GAAG,CAAC;SAC/B,KAAK,CAAC,KAAK,CAAC;SACZ,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;AACrD,CAAC;AAED;;;GAGG;AACH,SAAS,cAAc,CAAC,WAAqB,EAAE,YAAsB;IACnE,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAEpE,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,YAAY,CAAC,CAAC;IACxC,MAAM,OAAO,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IAE1D,gCAAgC;IAChC,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,WAAW,EAAE,GAAG,YAAY,CAAC,CAAC,CAAC;IACzD,OAAO,OAAO,CAAC,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC;AACrC,CAAC;AAED,iFAAiF;AAEjF;;;;;;;GAOG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,QAAgB,EAChB,aAAqB,EACrB,aAAqB,CAAC,EACtB,WAAmB,IAAI;IAEvB,MAAM,WAAW,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvC,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAExC,IAAI,QAAQ,CAAC;IACb,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,YAAY,EAAE,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,oEAAoE;IACpE,MAAM,gBAAgB,GAAG,QAAQ;SAC9B,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,aAAa,KAAK,aAAa,CAAC;SAC9C,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAEhB,MAAM,UAAU,GAAmB,EAAE,CAAC;IAEtC,KAAK,MAAM,WAAW,IAAI,gBAAgB,EAAE,CAAC;QAC3C,IAAI,OAA6B,CAAC;QAClC,IAAI,CAAC;YACH,OAAO,GAAG,MAAM,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC;QAC7C,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;QACD,IAAI,CAAC,OAAO,EAAE,QAAQ;YAAE,SAAS;QAEjC,+BAA+B;QAC/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YACrD,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAE,CAAC;YACjC,MAAM,IAAI,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;YAErC,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,IAAI,IAAI,EAAE,IAAI,KAAK,WAAW,EAAE,CAAC;gBACtD,kDAAkD;gBAClD,IAAI,GAAG,CAAC,OAAO,CAAC,UAAU,CAAC,eAAe,CAAC,IAAI,GAAG,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC;oBAAE,SAAS;gBAC3F,IAAI,GAAG,CAAC,OAAO,CAAC,UAAU,CAAC,iBAAiB,CAAC,IAAI,GAAG,CAAC,OAAO,CAAC,UAAU,CAAC,UAAU,CAAC;oBAAE,SAAS;gBAE9F,MAAM,cAAc,GAAG,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;gBAC7C,MAAM,KAAK,GAAG,cAAc,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;gBAE1D,IAAI,KAAK,IAAI,QAAQ,EAAE,CAAC;oBACtB,UAAU,CAAC,IAAI,CAAC;wBACd,QAAQ,EAAE,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;wBACnC,MAAM,EAAE,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,WAAW;wBAChD,SAAS,EAAE,OAAO,CAAC,EAAE;wBACrB,KAAK;qBACN,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,uCAAuC;IACvC,OAAO,UAAU;SACd,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC;SACjC,KAAK,CAAC,CAAC,EAAE,UAAU,CAAC,CAAC;AAC1B,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,CAAC,SAAyB;IAC3D,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAEtC,MAAM,KAAK,GAAG;QACZ,gCAAgC;QAChC,EAAE;QACF,4EAA4E;QAC5E,EAAE;KACH,CAAC;IAEF,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAClD,KAAK,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAChD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACjB,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAClB,KAAK,CAAC,IAAI,CAAC,iFAAiF,CAAC,CAAC;IAC9F,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEf,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}

package/dist/utils/tokens.d.ts

@@ -0,0 +1,53 @@
+/**
+ * Token estimation utilities for context window management.
+ *
+ * Uses a heuristic character-to-token ratio rather than a full tokenizer
+ * to keep dependencies minimal.  The ratios are calibrated for typical
+ * code / English text seen by llama-family models (~3.5–4 chars per token).
+ *
+ * The key concern is *budget management* — we don't need exact counts,
+ * just good-enough estimates to decide when to evict or summarize.
+ */
+/**
+ * Known context-window sizes (tokens) for popular local models.
+ * Used as fallback when the provider doesn't report a limit.
+ */
+export declare const MODEL_CONTEXT_LIMITS: Record<string, number>;
+/** Rough token count for a string. */
+export declare function estimateTokens(text: string): number;
+/** Estimate tokens for an array of messages (role + content). */
+export declare function estimateMessageTokens(messages: Array<{
+    role: string;
+    content: string;
+}>): number;
+/**
+ * Get the effective context budget for a model (in tokens).
+ * This is the max tokens we allow in the message array before
+ * triggering summarization / eviction.
+ */
+export declare function getContextBudget(model?: string): number;
+/**
+ * Check whether the current messages exceed the context budget.
+ * Returns { overBudget, currentTokens, budget, excess }.
+ */
+export declare function checkBudget(messages: Array<{
+    role: string;
+    content: string;
+}>, systemPrompt: string | undefined, model?: string): {
+    overBudget: boolean;
+    currentTokens: number;
+    budget: number;
+    excess: number;
+};
+/**
+ * Truncate text to fit within a token budget, keeping the beginning and end
+ * (most useful context is usually at boundaries).
+ */
+export declare function truncateToTokenBudget(text: string, maxTokens: number): string;
+/**
+ * Truncate tool output (file contents, search results) to a sensible size.
+ * - File reads: keep first + last N lines
+ * - Search results: keep first N matches
+ */
+export declare function truncateToolOutput(toolName: string, output: string, maxTokens?: number): string;
+//# sourceMappingURL=tokens.d.ts.map

package/dist/utils/tokens.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tokens.d.ts","sourceRoot":"","sources":["../../src/utils/tokens.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAOH;;;GAGG;AACH,eAAO,MAAM,oBAAoB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAgBvD,CAAC;AAaF,sCAAsC;AACtC,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAGnD;AAED,iEAAiE;AACjE,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,GAAG,MAAM,CAOhG;AAED;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,CAUvD;AAED;;;GAGG;AACH,wBAAgB,WAAW,CACzB,QAAQ,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,EAClD,YAAY,EAAE,MAAM,GAAG,SAAS,EAChC,KAAK,CAAC,EAAE,MAAM,GACb;IAAE,UAAU,EAAE,OAAO,CAAC;IAAC,aAAa,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CAahF;AAID;;;GAGG;AACH,wBAAgB,qBAAqB,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAY7E;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAChC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,MAAM,EACd,SAAS,GAAE,MAAa,GACvB,MAAM,CA2BR"}

package/dist/utils/tokens.js

@@ -0,0 +1,138 @@
+/**
+ * Token estimation utilities for context window management.
+ *
+ * Uses a heuristic character-to-token ratio rather than a full tokenizer
+ * to keep dependencies minimal.  The ratios are calibrated for typical
+ * code / English text seen by llama-family models (~3.5–4 chars per token).
+ *
+ * The key concern is *budget management* — we don't need exact counts,
+ * just good-enough estimates to decide when to evict or summarize.
+ */
+// ── Constants ─────────────────────────────────────────────────────────────────
+/** Average characters per token for English + code. */
+const CHARS_PER_TOKEN = 3.8;
+/**
+ * Known context-window sizes (tokens) for popular local models.
+ * Used as fallback when the provider doesn't report a limit.
+ */
+export const MODEL_CONTEXT_LIMITS = {
+    'llama3.2': 128_000,
+    'llama3.2:1b': 8_192,
+    'llama3.2:3b': 128_000,
+    'llama3.1': 128_000,
+    'llama3': 8_192,
+    'llama2': 4_096,
+    'mistral': 8_192,
+    'mixtral': 32_768,
+    'codellama': 16_384,
+    'deepseek-coder': 16_384,
+    'deepseek-coder-v2': 128_000,
+    'qwen2.5-coder': 128_000,
+    'phi3': 128_000,
+    'gemma2': 8_192,
+    'command-r': 128_000,
+};
+/** Safe default when model is unknown (conservative). */
+const DEFAULT_CONTEXT_LIMIT = 8_192;
+/**
+ * How much of the context window we're willing to fill with messages.
+ * Leave headroom for the model's own generation and system prompt.
+ */
+const USAGE_RATIO = 0.75;
+// ── Estimation functions ──────────────────────────────────────────────────────
+/** Rough token count for a string. */
+export function estimateTokens(text) {
+    if (!text)
+        return 0;
+    return Math.ceil(text.length / CHARS_PER_TOKEN);
+}
+/** Estimate tokens for an array of messages (role + content). */
+export function estimateMessageTokens(messages) {
+    let total = 0;
+    for (const msg of messages) {
+        // ~4 tokens overhead per message for role, delimiters
+        total += 4 + estimateTokens(msg.content);
+    }
+    return total;
+}
+/**
+ * Get the effective context budget for a model (in tokens).
+ * This is the max tokens we allow in the message array before
+ * triggering summarization / eviction.
+ */
+export function getContextBudget(model) {
+    if (!model)
+        return Math.floor(DEFAULT_CONTEXT_LIMIT * USAGE_RATIO);
+    // Try exact match first, then prefix match
+    const lower = model.toLowerCase();
+    const limit = MODEL_CONTEXT_LIMITS[lower]
+        ?? Object.entries(MODEL_CONTEXT_LIMITS).find(([k]) => lower.startsWith(k))?.[1]
+        ?? DEFAULT_CONTEXT_LIMIT;
+    return Math.floor(limit * USAGE_RATIO);
+}
+/**
+ * Check whether the current messages exceed the context budget.
+ * Returns { overBudget, currentTokens, budget, excess }.
+ */
+export function checkBudget(messages, systemPrompt, model) {
+    const budget = getContextBudget(model);
+    const systemTokens = systemPrompt ? estimateTokens(systemPrompt) + 4 : 0;
+    const msgTokens = estimateMessageTokens(messages);
+    const currentTokens = systemTokens + msgTokens;
+    const excess = currentTokens - budget;
+    return {
+        overBudget: excess > 0,
+        currentTokens,
+        budget,
+        excess: Math.max(0, excess),
+    };
+}
+// ── Text truncation helpers ───────────────────────────────────────────────────
+/**
+ * Truncate text to fit within a token budget, keeping the beginning and end
+ * (most useful context is usually at boundaries).
+ */
+export function truncateToTokenBudget(text, maxTokens) {
+    const estimated = estimateTokens(text);
+    if (estimated <= maxTokens)
+        return text;
+    const maxChars = Math.floor(maxTokens * CHARS_PER_TOKEN);
+    const keepChars = Math.floor(maxChars * 0.45); // 45% from start, 45% from end, ~10% for marker
+    const head = text.slice(0, keepChars);
+    const tail = text.slice(-keepChars);
+    const omittedTokens = estimated - maxTokens;
+    return `${head}\n\n[… ~${omittedTokens} tokens omitted …]\n\n${tail}`;
+}
+/**
+ * Truncate tool output (file contents, search results) to a sensible size.
+ * - File reads: keep first + last N lines
+ * - Search results: keep first N matches
+ */
+export function truncateToolOutput(toolName, output, maxTokens = 1500) {
+    const estimated = estimateTokens(output);
+    if (estimated <= maxTokens)
+        return output;
+    if (toolName === 'read_file') {
+        // For file reads, keep head + tail for best context
+        const lines = output.split('\n');
+        const maxLines = Math.floor(maxTokens / 10); // ~10 tokens per line avg
+        if (lines.length <= maxLines)
+            return output;
+        const keepLines = Math.floor(maxLines * 0.45);
+        const head = lines.slice(0, keepLines).join('\n');
+        const tail = lines.slice(-keepLines).join('\n');
+        const omitted = lines.length - keepLines * 2;
+        return `${head}\n\n[… ${omitted} lines omitted …]\n\n${tail}`;
+    }
+    if (toolName === 'search_text') {
+        // For search results, keep first N results (most relevant)
+        const lines = output.split('\n');
+        const maxLines = Math.floor(maxTokens / 8);
+        if (lines.length <= maxLines)
+            return output;
+        return lines.slice(0, maxLines).join('\n') + `\n\n[… ${lines.length - maxLines} more results truncated]`;
+    }
+    // Generic truncation
+    return truncateToTokenBudget(output, maxTokens);
+}
+//# sourceMappingURL=tokens.js.map

package/dist/utils/tokens.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"tokens.js","sourceRoot":"","sources":["../../src/utils/tokens.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,iFAAiF;AAEjF,uDAAuD;AACvD,MAAM,eAAe,GAAG,GAAG,CAAC;AAE5B;;;GAGG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAA2B;IAC1D,UAAU,EAAS,OAAO;IAC1B,aAAa,EAAQ,KAAK;IAC1B,aAAa,EAAM,OAAO;IAC1B,UAAU,EAAS,OAAO;IAC1B,QAAQ,EAAa,KAAK;IAC1B,QAAQ,EAAa,KAAK;IAC1B,SAAS,EAAY,KAAK;IAC1B,SAAS,EAAW,MAAM;IAC1B,WAAW,EAAS,MAAM;IAC1B,gBAAgB,EAAI,MAAM;IAC1B,mBAAmB,EAAC,OAAO;IAC3B,eAAe,EAAI,OAAO;IAC1B,MAAM,EAAc,OAAO;IAC3B,QAAQ,EAAc,KAAK;IAC3B,WAAW,EAAS,OAAO;CAC5B,CAAC;AAEF,yDAAyD;AACzD,MAAM,qBAAqB,GAAG,KAAK,CAAC;AAEpC;;;GAGG;AACH,MAAM,WAAW,GAAG,IAAI,CAAC;AAEzB,iFAAiF;AAEjF,sCAAsC;AACtC,MAAM,UAAU,cAAc,CAAC,IAAY;IACzC,IAAI,CAAC,IAAI;QAAE,OAAO,CAAC,CAAC;IACpB,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,eAAe,CAAC,CAAC;AAClD,CAAC;AAED,iEAAiE;AACjE,MAAM,UAAU,qBAAqB,CAAC,QAAkD;IACtF,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC3B,sDAAsD;QACtD,KAAK,IAAI,CAAC,GAAG,cAAc,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAC3C,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,KAAc;IAC7C,IAAI,CAAC,KAAK;QAAE,OAAO,IAAI,CAAC,KAAK,CAAC,qBAAqB,GAAG,WAAW,CAAC,CAAC;IAEnE,2CAA2C;IAC3C,MAAM,KAAK,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;IAClC,MAAM,KAAK,GAAG,oBAAoB,CAAC,KAAK,CAAC;WACpC,MAAM,CAAC,OAAO,CAAC,oBAAoB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;WAC5E,qBAAqB,CAAC;IAE3B,OAAO,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,WAAW,CAAC,CAAC;AACzC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,WAAW,CACzB,QAAkD,EAClD,YAAgC,EAChC,KAAc;IAEd,MAAM,MAAM,GAAG,gBAAgB,CAAC,KAAK,CAAC,CAAC;IACvC,MAAM,YAAY,GAAG,YAAY,CAAC,CAAC,CAAC,cAAc,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACzE,MAAM,SAAS,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;IAClD,MAAM,aAAa,GAAG,YAAY,GAAG,SAAS,CAAC;IAC/C,MAAM,MAAM,GAAG,aAAa,GAAG,MAAM,CAAC;IAEtC,OAAO;QACL,UAAU,EAAE,MAAM,GAAG,CAAC;QACtB,aAAa;QACb,MAAM;QACN,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC;KAC5B,CAAC;AACJ,CAAC;AAED,iFAAiF;AAEjF;;;GAGG;AACH,MAAM,UAAU,qBAAqB,CAAC,IAAY,EAAE,SAAiB;IACnE,MAAM,SAAS,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;IACvC,IAAI,SAAS,IAAI,SAAS;QAAE,OAAO,IAAI,CAAC;IAExC,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,eAAe,CAAC,CAAC;IACzD,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,GAAG,IAAI,CAAC,CAAC,CAAC,gDAAgD;IAE/F,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;IACtC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,SAAS,CAAC,CAAC;IACpC,MAAM,aAAa,GAAG,SAAS,GAAG,SAAS,CAAC;IAE5C,OAAO,GAAG,IAAI,WAAW,aAAa,yBAAyB,IAAI,EAAE,CAAC;AACxE,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,kBAAkB,CAChC,QAAgB,EAChB,MAAc,EACd,YAAoB,IAAI;IAExB,MAAM,SAAS,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IACzC,IAAI,SAAS,IAAI,SAAS;QAAE,OAAO,MAAM,CAAC;IAE1C,IAAI,QAAQ,KAAK,WAAW,EAAE,CAAC;QAC7B,oDAAoD;QACpD,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACjC,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,EAAE,CAAC,CAAC,CAAC,0BAA0B;QACvE,IAAI,KAAK,CAAC,MAAM,IAAI,QAAQ;YAAE,OAAO,MAAM,CAAC;QAE5C,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,GAAG,IAAI,CAAC,CAAC;QAC9C,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAClD,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,MAAM,OAAO,GAAG,KAAK,CAAC,MAAM,GAAG,SAAS,GAAG,CAAC,CAAC;QAC7C,OAAO,GAAG,IAAI,UAAU,OAAO,wBAAwB,IAAI,EAAE,CAAC;IAChE,CAAC;IAED,IAAI,QAAQ,KAAK,aAAa,EAAE,CAAC;QAC/B,2DAA2D;QAC3D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACjC,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC;QAC3C,IAAI,KAAK,CAAC,MAAM,IAAI,QAAQ;YAAE,OAAO,MAAM,CAAC;QAC5C,OAAO,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,UAAU,KAAK,CAAC,MAAM,GAAG,QAAQ,0BAA0B,CAAC;IAC3G,CAAC;IAED,qBAAqB;IACrB,OAAO,qBAAqB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;AAClD,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sunilp-org/jam-cli",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "Jam — developer-first AI assistant CLI for the terminal. Ask questions, explain code, review diffs, generate patches, and run agentic tasks powered by Ollama.",
   "type": "module",
   "license": "MIT",
@@ -23,6 +23,7 @@
     "assistant",
     "terminal",
     "ollama",
+    "embedded",
     "llm",
     "developer-tools",
     "code-review",
@@ -53,13 +54,16 @@
     "test:coverage": "vitest run --coverage",
     "clean": "rm -rf dist"
   },
+  "optionalDependencies": {
+    "keytar": "^7.9.0",
+    "node-llama-cpp": "^3.0.0"
+  },
   "dependencies": {
     "chalk": "^5.3.0",
     "commander": "^12.1.0",
     "cosmiconfig": "^9.0.0",
     "ink": "^5.0.1",
     "ink-text-input": "^6.0.0",
-    "keytar": "^7.9.0",
     "marked": "^12.0.0",
     "marked-terminal": "^7.1.0",
     "minimatch": "^9.0.5",