diff --git a/apps/web/src/app/api/agents/[id]/conversation/[conversationId]/route.ts b/apps/web/src/app/api/agents/[id]/conversation/[conversationId]/route.ts
new file mode 100644
index 0000000..ad638b1
--- /dev/null
+++ b/apps/web/src/app/api/agents/[id]/conversation/[conversationId]/route.ts
@@ -0,0 +1,601 @@
+/**
+ * Agent Conversation Endpoint (ID-based version)
+ *
+ * POST: Send a message and stream the AI response
+ * GET: Retrieve conversation history
+ * DELETE: Delete a conversation
+ *
+ * This endpoint uses agent id directly for dashboard usage
+ */
+
+import { decryptApiKey } from '@/lib/crypto/api-keys';
+import { Prisma, prisma } from '@tpmjs/db';
+import type { AIProvider } from '@tpmjs/types/agent';
+import { SendMessageSchema } from '@tpmjs/types/agent';
+import type { LanguageModel, ModelMessage } from 'ai';
+import { type NextRequest, NextResponse } from 'next/server';
+import { type RateLimitConfig, checkRateLimitDistributed } from '~/lib/rate-limit';
+
+/**
+ * Rate limit for chat messages: 30 requests per minute
+ * This is stricter than default because chat involves expensive LLM calls
+ */
+const CHAT_RATE_LIMIT: RateLimitConfig = {
+  limit: 30,
+  windowSeconds: 60,
+};
+
+export const runtime = 'nodejs';
+export const dynamic = 'force-dynamic';
+export const maxDuration = 300; // 5 minutes for long agentic runs
+
+type RouteContext = {
+  params: Promise<{ id: string; conversationId: string }>;
+};
+
+/**
+ * Get AI provider SDK based on provider type
+ */
+async function getProviderModel(
+  provider: AIProvider,
+  modelId: string,
+  apiKey: string
+): Promise<LanguageModel> {
+  switch (provider) {
+    case 'OPENAI': {
+      const { createOpenAI } = await import('@ai-sdk/openai');
+      return createOpenAI({ apiKey })(modelId);
+    }
+    case 'ANTHROPIC': {
+      const { createAnthropic } = await import('@ai-sdk/anthropic');
+      return createAnthropic({ apiKey })(modelId);
+    }
+    case 'GOOGLE': {
+      const { createGoogleGenerativeAI } = await import('@ai-sdk/google');
+      return createGoogleGenerativeAI({ apiKey })(modelId);
+    }
+    case 'GROQ': {
+      const { createGroq } = await import('@ai-sdk/groq');
+      return createGroq({ apiKey })(modelId);
+    }
+    case 'MISTRAL': {
+      const { createMistral } = await import('@ai-sdk/mistral');
+      return createMistral({ apiKey })(modelId);
+    }
+    default:
+      throw new Error(`Unsupported provider: ${provider}`);
+  }
+}
+
+/**
+ * POST /api/agents/[id]/conversation/[conversationId]
+ * Send a message and stream the AI response via SSE
+ */
+export async function POST(request: NextRequest, context: RouteContext): Promise<Response> {
+  // Check rate limit first to prevent expensive LLM calls (uses distributed KV when available)
+  const rateLimitResponse = await checkRateLimitDistributed(request, CHAT_RATE_LIMIT);
+  if (rateLimitResponse) {
+    return rateLimitResponse;
+  }
+
+  const { id: agentId, conversationId } = await context.params;
+
+  try {
+    const body = await request.json();
+    const parsed = SendMessageSchema.safeParse(body);
+    if (!parsed.success) {
+      return NextResponse.json(
+        { success: false, error: 'Invalid request', details: parsed.error.flatten() },
+        { status: 400 }
+      );
+    }
+
+    // Fetch agent with all tool relations using agent ID
+    const { fetchAgentWithTools, buildAgentTools } = await import('@/lib/agents/build-tools');
+    const agent = await fetchAgentWithTools(agentId);
+
+    if (!agent) {
+      return NextResponse.json({ success: false, error: 'Agent not found' }, { status: 404 });
+    }
+
+    // Map provider to expected key name format
+    const providerKeyNames: Record<string, string> = {
+      OPENAI: 'OPENAI_API_KEY',
+      ANTHROPIC: 'ANTHROPIC_API_KEY',
+      GOOGLE: 'GOOGLE_API_KEY',
+      GROQ: 'GROQ_API_KEY',
+      MISTRAL: 'MISTRAL_API_KEY',
+    };
+    const keyName = providerKeyNames[agent.provider];
+
+    if (!keyName) {
+      return NextResponse.json(
+        { success: false, error: `Unsupported provider: ${agent.provider}` },
+        { status: 400 }
+      );
+    }
+
+    // Get user's API key for this provider
+    const userApiKey = await prisma.userApiKey.findUnique({
+      where: {
+        userId_keyName: {
+          userId: agent.userId,
+          keyName,
+        },
+      },
+    });
+
+    if (!userApiKey) {
+      return NextResponse.json(
+        {
+          success: false,
+          error: `No API key configured for ${agent.provider}. Please add your API key in settings.`,
+        },
+        { status: 400 }
+      );
+    }
+
+    // Decrypt the API key
+    const apiKey = decryptApiKey(userApiKey.encryptedKey, userApiKey.keyIv);
+
+    // Get or create conversation
+    let conversation = await prisma.conversation.findUnique({
+      where: {
+        agentId_slug: {
+          agentId: agent.id,
+          slug: conversationId,
+        },
+      },
+    });
+
+    if (!conversation) {
+      conversation = await prisma.conversation.create({
+        data: {
+          agentId: agent.id,
+          slug: conversationId,
+          title: parsed.data.message.slice(0, 100),
+        },
+      });
+    }
+
+    // Fetch recent messages for context
+    const recentMessages = await prisma.message.findMany({
+      where: { conversationId: conversation.id },
+      orderBy: { createdAt: 'desc' },
+      take: agent.maxMessagesInContext,
+    });
+
+    // Reverse to get chronological order
+    recentMessages.reverse();
+
+    // Save user message
+    await prisma.message.create({
+      data: {
+        conversationId: conversation.id,
+        role: 'USER',
+        content: parsed.data.message,
+      },
+    });
+
+    // Build AI SDK messages from conversation history
+    const { streamText, stepCountIs } = await import('ai');
+
+    const messages: ModelMessage[] = [];
+
+    // Add system prompt if defined
+    if (agent.systemPrompt) {
+      messages.push({
+        role: 'system',
+        content: agent.systemPrompt,
+      });
+    }
+
+    // Add conversation history - properly format for AI SDK
+    for (const msg of recentMessages) {
+      if (msg.role === 'USER') {
+        messages.push({ role: 'user', content: msg.content });
+      } else if (msg.role === 'ASSISTANT') {
+        // For assistant messages with tool calls, include ToolCallParts in content
+        if (msg.toolCalls && Array.isArray(msg.toolCalls) && msg.toolCalls.length > 0) {
+          const toolCallParts = (
+            msg.toolCalls as Array<{ toolCallId: string; toolName: string; args: unknown }>
+          ).map((tc) => ({
+            type: 'tool-call' as const,
+            toolCallId: tc.toolCallId,
+            toolName: tc.toolName,
+            input: tc.args,
+          }));
+          // Content includes text (if any) plus tool call parts
+          const content: Array<
+            | { type: 'text'; text: string }
+            | { type: 'tool-call'; toolCallId: string; toolName: string; input: unknown }
+          > = [];
+          if (msg.content) {
+            content.push({ type: 'text', text: msg.content });
+          }
+          content.push(...toolCallParts);
+          messages.push({
+            role: 'assistant',
+            content,
+          });
+        } else {
+          messages.push({
+            role: 'assistant',
+            content: msg.content,
+          });
+        }
+      } else if (msg.role === 'TOOL') {
+        // Tool results use the 'tool' role with content array
+        messages.push({
+          role: 'tool',
+          content: [
+            {
+              type: 'tool-result' as const,
+              toolCallId: msg.toolCallId || '',
+              toolName: msg.toolName || '',
+              output: {
+                type: 'json' as const,
+                value: msg.toolResult,
+              },
+            },
+          ],
+        });
+      }
+    }
+
+    // Add new user message
+    messages.push({ role: 'user', content: parsed.data.message });
+
+    // Build tools from agent configuration
+    const tools = buildAgentTools(agent);
+
+    // Get the provider model
+    const model = await getProviderModel(agent.provider, agent.modelId, apiKey);
+
+    // Create SSE stream
+    const stream = new ReadableStream({
+      async start(controller) {
+        const encoder = new TextEncoder();
+
+        const sendEvent = (event: string, data: unknown) => {
+          const message = `event: ${event}\ndata: ${JSON.stringify(data)}\n\n`;
+          controller.enqueue(encoder.encode(message));
+        };
+
+        try {
+          const startTime = Date.now();
+          let fullContent = '';
+          // Accumulate tool calls with their input args
+          const toolCallsMap: Map<string, { toolCallId: string; toolName: string; args: unknown }> =
+            new Map();
+          // Collect tool results to save AFTER assistant message (for correct chronological order)
+          const pendingToolResults: Array<{
+            toolCallId: string;
+            toolName: string;
+            output: unknown;
+          }> = [];
+          let inputTokens = 0;
+          let outputTokens = 0;
+
+          // Stream the response with agentic loop control
+          const result = await streamText({
+            model,
+            messages,
+            tools,
+            stopWhen: stepCountIs(agent.maxToolCallsPerTurn),
+            onChunk: async ({ chunk }) => {
+              // Stream tool calls as they come in and capture their inputs
+              if (chunk.type === 'tool-call') {
+                const input = 'args' in chunk ? chunk.args : chunk.input;
+                // Store tool call with input for later persistence
+                toolCallsMap.set(chunk.toolCallId, {
+                  toolCallId: chunk.toolCallId,
+                  toolName: chunk.toolName,
+                  args: input,
+                });
+                sendEvent('tool_call', {
+                  toolCallId: chunk.toolCallId,
+                  toolName: chunk.toolName,
+                  input,
+                });
+              }
+            },
+            onStepFinish: async ({ toolCalls, toolResults, usage }) => {
+              // Capture tool calls from step finish (backup in case onChunk missed any)
+              if (toolCalls && Array.isArray(toolCalls)) {
+                for (const tc of toolCalls) {
+                  if (!toolCallsMap.has(tc.toolCallId)) {
+                    // Use 'input' from DynamicToolCall or fall back to type assertion for typed calls
+                    const args =
+                      'input' in tc ? tc.input : 'args' in tc ? (tc as { args: unknown }).args : {};
+                    toolCallsMap.set(tc.toolCallId, {
+                      toolCallId: tc.toolCallId,
+                      toolName: tc.toolName,
+                      args,
+                    });
+                  }
+                }
+              }
+
+              // Send tool results via SSE but don't save yet (save after assistant message for correct order)
+              if (toolResults && toolResults.length > 0) {
+                for (const tr of toolResults) {
+                  sendEvent('tool_result', {
+                    toolCallId: tr.toolCallId,
+                    output: tr.output,
+                  });
+
+                  // Collect tool results to save after assistant message
+                  pendingToolResults.push({
+                    toolCallId: tr.toolCallId,
+                    toolName: tr.toolName,
+                    output: tr.output,
+                  });
+                }
+              }
+
+              // Track token usage
+              if (usage) {
+                inputTokens += usage.inputTokens ?? 0;
+                outputTokens += usage.outputTokens ?? 0;
+              }
+            },
+          });
+
+          // Stream text chunks
+          for await (const chunk of result.textStream) {
+            fullContent += chunk;
+            sendEvent('chunk', { type: 'text', text: chunk });
+          }
+
+          // Get final response data
+          const finalUsage = await result.usage;
+
+          // Convert tool calls map to array for storage
+          const allToolCalls = Array.from(toolCallsMap.values());
+
+          // Update token counts from final usage
+          if (finalUsage) {
+            inputTokens = finalUsage.inputTokens ?? inputTokens;
+            outputTokens = finalUsage.outputTokens ?? outputTokens;
+          }
+
+          // Save assistant message FIRST (so it has earlier createdAt than tool results)
+          const assistantMessage = await prisma.message.create({
+            data: {
+              conversationId: conversation.id,
+              role: 'ASSISTANT',
+              content: fullContent,
+              // Cast to Prisma-compatible JSON type
+              toolCalls:
+                allToolCalls.length > 0
+                  ? (allToolCalls as unknown as Prisma.InputJsonValue)
+                  : Prisma.JsonNull,
+              inputTokens,
+              outputTokens,
+            },
+          });
+
+          // Now save TOOL messages (after assistant, for correct chronological order)
+          for (const tr of pendingToolResults) {
+            await prisma.message.create({
+              data: {
+                conversationId: conversation.id,
+                role: 'TOOL',
+                content: JSON.stringify(tr.output),
+                toolCallId: tr.toolCallId,
+                toolName: tr.toolName,
+                toolResult: tr.output as object,
+              },
+            });
+          }
+
+          // Update conversation timestamp
+          await prisma.conversation.update({
+            where: { id: conversation.id },
+            data: { updatedAt: new Date() },
+          });
+
+          const executionTimeMs = Date.now() - startTime;
+
+          // Send token usage
+          sendEvent('tokens', {
+            inputTokens,
+            outputTokens,
+            totalTokens: inputTokens + outputTokens,
+          });
+
+          // Send completion event
+          sendEvent('complete', {
+            messageId: assistantMessage.id,
+            conversationId: conversation.id,
+            executionTimeMs,
+          });
+        } catch (error) {
+          console.error('Agent conversation error:', error);
+          sendEvent('error', {
+            message: error instanceof Error ? error.message : 'Unknown error',
+          });
+        } finally {
+          controller.close();
+        }
+      },
+    });
+
+    return new NextResponse(stream, {
+      headers: {
+        'Content-Type': 'text/event-stream',
+        'Cache-Control': 'no-cache',
+        Connection: 'keep-alive',
+      },
+    });
+  } catch (error) {
+    console.error('Failed to process message:', error);
+    return NextResponse.json(
+      {
+        success: false,
+        error: error instanceof Error ? error.message : 'Failed to process message',
+      },
+      { status: 500 }
+    );
+  }
+}
+
+/**
+ * GET /api/agents/[id]/conversation/[conversationId]
+ * Retrieve conversation history with pagination
+ *
+ * Query params:
+ * - limit: Max messages to return (default: 50, max: 100)
+ * - before: Fetch messages created before this ISO timestamp (for loading older messages)
+ * - after: Fetch messages created after this ISO timestamp (for loading newer messages)
+ *
+ * Default behavior (no before/after): Returns the most recent messages
+ * With before: Returns messages older than the timestamp (for scrolling up)
+ * With after: Returns messages newer than the timestamp (for refreshing)
+ */
+export async function GET(request: NextRequest, context: RouteContext): Promise<NextResponse> {
+  const { id: agentId, conversationId } = await context.params;
+  const { searchParams } = new URL(request.url);
+
+  const limit = Math.min(Number.parseInt(searchParams.get('limit') || '50', 10), 100);
+  const before = searchParams.get('before');
+  const after = searchParams.get('after');
+
+  try {
+    // Fetch agent by ID
+    const agent = await prisma.agent.findUnique({
+      where: { id: agentId },
+      select: { id: true },
+    });
+
+    if (!agent) {
+      return NextResponse.json({ success: false, error: 'Agent not found' }, { status: 404 });
+    }
+
+    // Fetch conversation
+    const conversation = await prisma.conversation.findUnique({
+      where: {
+        agentId_slug: {
+          agentId: agent.id,
+          slug: conversationId,
+        },
+      },
+    });
+
+    if (!conversation) {
+      return NextResponse.json(
+        { success: false, error: 'Conversation not found' },
+        { status: 404 }
+      );
+    }
+
+    // Build the where clause based on cursor
+    const whereClause: {
+      conversationId: string;
+      createdAt?: { lt?: Date; gt?: Date };
+    } = { conversationId: conversation.id };
+
+    if (before) {
+      whereClause.createdAt = { lt: new Date(before) };
+    } else if (after) {
+      whereClause.createdAt = { gt: new Date(after) };
+    }
+
+    // Determine fetch order:
+    // - Default (no cursor) or "before": Fetch desc (newest first), then reverse for chronological order
+    // - "after": Fetch asc (oldest first) to get messages after the cursor
+    const shouldFetchDesc = !after;
+
+    // Fetch messages
+    const messages = await prisma.message.findMany({
+      where: whereClause,
+      orderBy: { createdAt: shouldFetchDesc ? 'desc' : 'asc' },
+      take: limit + 1,
+    });
+
+    const hasMore = messages.length > limit;
+    let paginatedMessages = hasMore ? messages.slice(0, limit) : messages;
+
+    // Reverse if we fetched in desc order to maintain chronological order
+    if (shouldFetchDesc) {
+      paginatedMessages = paginatedMessages.reverse();
+    }
+
+    const mappedMessages = paginatedMessages.map((m) => ({
+      id: m.id,
+      role: m.role,
+      content: m.content,
+      toolCalls: m.toolCalls,
+      toolCallId: m.toolCallId,
+      toolName: m.toolName,
+      toolResult: m.toolResult,
+      inputTokens: m.inputTokens,
+      outputTokens: m.outputTokens,
+      createdAt: m.createdAt,
+    }));
+
+    return NextResponse.json({
+      success: true,
+      data: {
+        id: conversation.id,
+        slug: conversation.slug,
+        title: conversation.title,
+        createdAt: conversation.createdAt,
+        updatedAt: conversation.updatedAt,
+        messages: mappedMessages,
+      },
+      pagination: {
+        limit,
+        hasMore,
+        ...(before && { before }),
+        ...(after && { after }),
+      },
+    });
+  } catch (error) {
+    console.error('Failed to fetch conversation:', error);
+    return NextResponse.json(
+      { success: false, error: 'Failed to fetch conversation' },
+      { status: 500 }
+    );
+  }
+}
+
+/**
+ * DELETE /api/agents/[id]/conversation/[conversationId]
+ * Delete a conversation
+ */
+export async function DELETE(_request: NextRequest, context: RouteContext): Promise<NextResponse> {
+  const { id: agentId, conversationId } = await context.params;
+
+  try {
+    // Fetch agent by ID
+    const agent = await prisma.agent.findUnique({
+      where: { id: agentId },
+      select: { id: true },
+    });
+
+    if (!agent) {
+      return NextResponse.json({ success: false, error: 'Agent not found' }, { status: 404 });
+    }
+
+    // Delete conversation (messages cascade)
+    await prisma.conversation.deleteMany({
+      where: {
+        agentId: agent.id,
+        slug: conversationId,
+      },
+    });
+
+    return NextResponse.json({
+      success: true,
+      data: { deleted: true },
+    });
+  } catch (error) {
+    console.error('Failed to delete conversation:', error);
+    return NextResponse.json(
+      { success: false, error: 'Failed to delete conversation' },
+      { status: 500 }
+    );
+  }
+}
diff --git a/apps/web/src/app/dashboard/agents/[id]/chat/[chatId]/page.tsx b/apps/web/src/app/dashboard/agents/[id]/chat/[chatId]/page.tsx
index 870ca84..d1877a8 100644
--- a/apps/web/src/app/dashboard/agents/[id]/chat/[chatId]/page.tsx
+++ b/apps/web/src/app/dashboard/agents/[id]/chat/[chatId]/page.tsx
@@ -225,7 +225,7 @@ export default function AgentChatPage(): React.ReactElement {
     if (!agent) return;
 
     try {
-      const response = await fetch(`/api/agents/${agent.uid}/conversations`);
+      const response = await fetch(`/api/agents/${agentId}/conversations`);
       const data = await response.json();
 
       if (data.success) {
@@ -234,14 +234,14 @@ export default function AgentChatPage(): React.ReactElement {
     } catch (err) {
       console.error('Failed to fetch conversations:', err);
     }
-  }, [agent]);
+  }, [agent, agentId]);
 
   // Fetch messages for active conversation
   const fetchMessages = useCallback(async () => {
     if (!agent || !activeConversationId) return;
 
     try {
-      const response = await fetch(`/api/agents/${agent.uid}/conversation/${activeConversationId}`);
+      const response = await fetch(`/api/agents/${agentId}/conversation/${activeConversationId}`);
       const data = await response.json();
 
       if (data.success) {
@@ -256,7 +256,7 @@ export default function AgentChatPage(): React.ReactElement {
     } catch (err) {
       console.error('Failed to fetch messages:', err);
     }
-  }, [agent, activeConversationId]);
+  }, [agent, agentId, activeConversationId]);
 
   useEffect(() => {
     const init = async () => {
@@ -312,7 +312,7 @@ export default function AgentChatPage(): React.ReactElement {
     setMessages((prev) => [...prev, userMessage]);
 
     try {
-      const response = await fetch(`/api/agents/${agent.uid}/conversation/${conversationId}`, {
+      const response = await fetch(`/api/agents/${agentId}/conversation/${conversationId}`, {
         method: 'POST',
         headers: { 'Content-Type': 'application/json' },
         body: JSON.stringify({ message: messageContent }),