nemoris 0.1.0 → 0.1.2

package/src/runtime/session-compactor.js

@@ -1,182 +1,182 @@
-import { ProviderRegistry } from "../providers/registry.js";
-
-const DEFAULT_MAX_TURNS = 8;
-const DEFAULT_KEEP_RECENT_TURNS = 4;
-const DEFAULT_COMPACTION_MODEL = "ollama/qwen3:8b";
-
-const SUMMARISE_PROMPT =
-  "Summarise the key facts, decisions, and context from this conversation history in 3-5 sentences. " +
-  "Preserve any specific values, filenames, decisions, or action items mentioned.";
-
-const CONDENSE_PROMPT =
-  "Condense these conversation summaries into a single high-level summary. " +
-  "Focus on long-term goals and key outcomes.";
-
-/**
- * Compact a turns array by summarising old turns into a single context message.
- *
- * @param {Array} turns - Array of {role, content, timestamp} objects
- * @param {object} options
- * @param {number} [options.maxTurns=8] - Trigger compaction when turns.length >= maxTurns
- * @param {number} [options.keepRecentTurns=4] - Always keep the last N turns verbatim
- * @param {string} [options.compactionModel] - Model to use for summarisation
- * @param {object} [options.registry] - ProviderRegistry instance (for testing)
- * @param {object} [options.ledger] - ContextLedger instance
- * @param {string} [options.sessionId] - Session ID
- * @param {number} [options.condensedFanout=4] - Number of depth-0 summaries before depth-1 condensation
- * @returns {{ compacted: boolean, turns: Array, summary?: string }}
- */
-export async function compactSessionContext(turns, options = {}) {
-  const maxTurns = options.maxTurns ?? DEFAULT_MAX_TURNS;
-  const keepRecentTurns = options.keepRecentTurns ?? DEFAULT_KEEP_RECENT_TURNS;
-  const compactionModel = options.compactionModel ?? DEFAULT_COMPACTION_MODEL;
-  const condensedFanout = options.condensedFanout ?? 4;
-  const ledger = options.ledger;
-  const sessionId = options.sessionId;
-
-  if (!Array.isArray(turns) || turns.length < maxTurns) {
-    return { compacted: false, turns };
-  }
-
-  const oldTurns = turns.slice(0, turns.length - keepRecentTurns);
-  const recentTurns = turns.slice(-keepRecentTurns);
-
-  try {
-    const registry = options.registry ?? new ProviderRegistry();
-    const adapter = getAdapter(registry, compactionModel);
-    const modelName = getModelName(compactionModel);
-
-    const historyText = JSON.stringify(oldTurns, null, 2);
-    const payload = {
-      model: modelName,
-      system: SUMMARISE_PROMPT,
-      messages: [{ role: "user", content: historyText }],
-    };
-
-    let summary;
-    try {
-      const raw = await adapter.invoke(payload);
-      const normalized = adapter.normalizeResponse(raw);
-      summary = normalized.output || normalized.summary || "";
-      
-      if (!summary || summary.length > 2000) {
-        throw new Error(summary.length > 2000 ? "Summary too long" : "Empty summary");
-      }
-    } catch (err) {
-      summary = level3Fallback(oldTurns, sessionId);
-    }
-
-    // DAG integration
-    if (ledger && sessionId) {
-      try {
-        ledger.saveContextSummary({
-          session_id: sessionId,
-          depth: 0,
-          source_event_ids: "[]", // We don't have individual event IDs here
-          summary_text: summary,
-          token_count: 0,
-        });
-
-        const depth0s = ledger.getContextSummaries({ session_id: sessionId, depth: 0 });
-        if (depth0s.length >= condensedFanout) {
-          await runDepth1Condensation(ledger, sessionId, depth0s, { adapter, modelName, condensedFanout });
-        }
-      } catch (dagErr) {
-        console.error(JSON.stringify({ service: "session_compactor", event: "dag_error", error: dagErr.message }));
-      }
-    }
-
-    return {
-      compacted: true,
-      turns: [
-        {
-          role: "assistant",
-          content: `[Context summary] ${summary}`,
-          timestamp: new Date().toISOString(),
-          compacted: true,
-        },
-        ...recentTurns,
-      ],
-      summary,
-    };
-  } catch (err) {
-    console.error(
-      JSON.stringify({ service: "session_compactor", error: err.message })
-    );
-    return { compacted: false, turns };
-  }
-}
-
-function getAdapter(registry, compactionModel) {
-  const [providerPrefix] = compactionModel.split("/");
-  let providerConfig;
-  if (providerPrefix === "ollama") {
-    providerConfig = {
-      id: "ollama",
-      baseUrl: process.env.OLLAMA_BASE_URL || "http://localhost:11434",
-    };
-  } else if (providerPrefix === "openrouter") {
-    providerConfig = {
-      id: "openrouter",
-      adapter: "openrouter",
-      baseUrl: "https://openrouter.ai/api/v1",
-      authRef: "env:OPENROUTER_API_KEY",
-    };
-  } else {
-    providerConfig = {
-      id: providerPrefix,
-      adapter: providerPrefix,
-      baseUrl: process.env.ANTHROPIC_BASE_URL || "https://api.anthropic.com",
-      authRef: "env:ANTHROPIC_API_KEY",
-    };
-  }
-  return registry.create(providerConfig);
-}
-
-function getModelName(compactionModel) {
-  const [, ...rest] = compactionModel.split("/");
-  return rest.join("/") || compactionModel;
-}
-
-function level3Fallback(oldTurns, sessionId) {
-  const raw = JSON.stringify(oldTurns);
-  const truncated = raw.slice(0, 1500);
-  const summary = `${truncated} [compacted — full history in context_events]`;
-  console.warn(`[session-compactor] Level 3 fallback triggered for session ${sessionId || "unknown"}`);
-  return summary;
-}
-
-async function runDepth1Condensation(ledger, sessionId, depth0s, { adapter, modelName }) {
-  const combinedText = depth0s.map(s => s.summary_text).join("\n---\n");
-  
-  let condensed;
-  try {
-    const payload = {
-      model: modelName,
-      system: CONDENSE_PROMPT,
-      messages: [{ role: "user", content: combinedText }],
-    };
-    const raw = await adapter.invoke(payload);
-    const normalized = adapter.normalizeResponse(raw);
-    condensed = normalized.output || normalized.summary || "";
-    
-    if (!condensed || condensed.length > 2000) {
-      throw new Error("Depth-1 summary failed or too long");
-    }
-  } catch (err) {
-    // Depth-1 fallback: concatenate first 500 chars of each depth-0 summary
-    condensed = depth0s.map(s => s.summary_text.slice(0, 500)).join("\n---\n");
-    console.warn(`[session-compactor] Level 3 fallback triggered for depth-1 session ${sessionId}`);
-  }
-
-  ledger.saveContextSummary({
-    session_id: sessionId,
-    depth: 1,
-    source_event_ids: JSON.stringify(depth0s.map(s => s.id)),
-    summary_text: condensed,
-    token_count: 0,
-  });
-
-  // Delete source depth-0 summaries
-  ledger.deleteContextSummaries({ session_id: sessionId, depth: 0 });
-}
+import { ProviderRegistry } from "../providers/registry.js";
+
+const DEFAULT_MAX_TURNS = 8;
+const DEFAULT_KEEP_RECENT_TURNS = 4;
+const DEFAULT_COMPACTION_MODEL = "ollama/qwen3:8b";
+
+const SUMMARISE_PROMPT =
+  "Summarise the key facts, decisions, and context from this conversation history in 3-5 sentences. " +
+  "Preserve any specific values, filenames, decisions, or action items mentioned.";
+
+const CONDENSE_PROMPT =
+  "Condense these conversation summaries into a single high-level summary. " +
+  "Focus on long-term goals and key outcomes.";
+
+/**
+ * Compact a turns array by summarising old turns into a single context message.
+ *
+ * @param {Array} turns - Array of {role, content, timestamp} objects
+ * @param {object} options
+ * @param {number} [options.maxTurns=8] - Trigger compaction when turns.length >= maxTurns
+ * @param {number} [options.keepRecentTurns=4] - Always keep the last N turns verbatim
+ * @param {string} [options.compactionModel] - Model to use for summarisation
+ * @param {object} [options.registry] - ProviderRegistry instance (for testing)
+ * @param {object} [options.ledger] - ContextLedger instance
+ * @param {string} [options.sessionId] - Session ID
+ * @param {number} [options.condensedFanout=4] - Number of depth-0 summaries before depth-1 condensation
+ * @returns {{ compacted: boolean, turns: Array, summary?: string }}
+ */
+export async function compactSessionContext(turns, options = {}) {
+  const maxTurns = options.maxTurns ?? DEFAULT_MAX_TURNS;
+  const keepRecentTurns = options.keepRecentTurns ?? DEFAULT_KEEP_RECENT_TURNS;
+  const compactionModel = options.compactionModel ?? DEFAULT_COMPACTION_MODEL;
+  const condensedFanout = options.condensedFanout ?? 4;
+  const ledger = options.ledger;
+  const sessionId = options.sessionId;
+
+  if (!Array.isArray(turns) || turns.length < maxTurns) {
+    return { compacted: false, turns };
+  }
+
+  const oldTurns = turns.slice(0, turns.length - keepRecentTurns);
+  const recentTurns = turns.slice(-keepRecentTurns);
+
+  try {
+    const registry = options.registry ?? new ProviderRegistry();
+    const adapter = getAdapter(registry, compactionModel);
+    const modelName = getModelName(compactionModel);
+
+    const historyText = JSON.stringify(oldTurns, null, 2);
+    const payload = {
+      model: modelName,
+      system: SUMMARISE_PROMPT,
+      messages: [{ role: "user", content: historyText }],
+    };
+
+    let summary;
+    try {
+      const raw = await adapter.invoke(payload);
+      const normalized = adapter.normalizeResponse(raw);
+      summary = normalized.output || normalized.summary || "";
+      
+      if (!summary || summary.length > 2000) {
+        throw new Error(summary.length > 2000 ? "Summary too long" : "Empty summary");
+      }
+    } catch (err) {
+      summary = level3Fallback(oldTurns, sessionId);
+    }
+
+    // DAG integration
+    if (ledger && sessionId) {
+      try {
+        ledger.saveContextSummary({
+          session_id: sessionId,
+          depth: 0,
+          source_event_ids: "[]", // We don't have individual event IDs here
+          summary_text: summary,
+          token_count: 0,
+        });
+
+        const depth0s = ledger.getContextSummaries({ session_id: sessionId, depth: 0 });
+        if (depth0s.length >= condensedFanout) {
+          await runDepth1Condensation(ledger, sessionId, depth0s, { adapter, modelName, condensedFanout });
+        }
+      } catch (dagErr) {
+        console.error(JSON.stringify({ service: "session_compactor", event: "dag_error", error: dagErr.message }));
+      }
+    }
+
+    return {
+      compacted: true,
+      turns: [
+        {
+          role: "assistant",
+          content: `[Context summary] ${summary}`,
+          timestamp: new Date().toISOString(),
+          compacted: true,
+        },
+        ...recentTurns,
+      ],
+      summary,
+    };
+  } catch (err) {
+    console.error(
+      JSON.stringify({ service: "session_compactor", error: err.message })
+    );
+    return { compacted: false, turns };
+  }
+}
+
+function getAdapter(registry, compactionModel) {
+  const [providerPrefix] = compactionModel.split("/");
+  let providerConfig;
+  if (providerPrefix === "ollama") {
+    providerConfig = {
+      id: "ollama",
+      baseUrl: process.env.OLLAMA_BASE_URL || "http://localhost:11434",
+    };
+  } else if (providerPrefix === "openrouter") {
+    providerConfig = {
+      id: "openrouter",
+      adapter: "openrouter",
+      baseUrl: "https://openrouter.ai/api/v1",
+      authRef: "env:OPENROUTER_API_KEY",
+    };
+  } else {
+    providerConfig = {
+      id: providerPrefix,
+      adapter: providerPrefix,
+      baseUrl: process.env.ANTHROPIC_BASE_URL || "https://api.anthropic.com",
+      authRef: "env:ANTHROPIC_API_KEY",
+    };
+  }
+  return registry.create(providerConfig);
+}
+
+function getModelName(compactionModel) {
+  const [, ...rest] = compactionModel.split("/");
+  return rest.join("/") || compactionModel;
+}
+
+function level3Fallback(oldTurns, sessionId) {
+  const raw = JSON.stringify(oldTurns);
+  const truncated = raw.slice(0, 1500);
+  const summary = `${truncated} [compacted — full history in context_events]`;
+  console.warn(`[session-compactor] Level 3 fallback triggered for session ${sessionId || "unknown"}`);
+  return summary;
+}
+
+async function runDepth1Condensation(ledger, sessionId, depth0s, { adapter, modelName }) {
+  const combinedText = depth0s.map(s => s.summary_text).join("\n---\n");
+  
+  let condensed;
+  try {
+    const payload = {
+      model: modelName,
+      system: CONDENSE_PROMPT,
+      messages: [{ role: "user", content: combinedText }],
+    };
+    const raw = await adapter.invoke(payload);
+    const normalized = adapter.normalizeResponse(raw);
+    condensed = normalized.output || normalized.summary || "";
+    
+    if (!condensed || condensed.length > 2000) {
+      throw new Error("Depth-1 summary failed or too long");
+    }
+  } catch (err) {
+    // Depth-1 fallback: concatenate first 500 chars of each depth-0 summary
+    condensed = depth0s.map(s => s.summary_text.slice(0, 500)).join("\n---\n");
+    console.warn(`[session-compactor] Level 3 fallback triggered for depth-1 session ${sessionId}`);
+  }
+
+  ledger.saveContextSummary({
+    session_id: sessionId,
+    depth: 1,
+    source_event_ids: JSON.stringify(depth0s.map(s => s.id)),
+    summary_text: condensed,
+    token_count: 0,
+  });
+
+  // Delete source depth-0 summaries
+  ledger.deleteContextSummaries({ session_id: sessionId, depth: 0 });
+}