diff --git a/.opencode/opencode.jsonc b/.opencode/opencode.jsonc
index dc0bee7c3f..88adf3762b 100644
--- a/.opencode/opencode.jsonc
+++ b/.opencode/opencode.jsonc
@@ -10,17 +10,5 @@
       "options": {},
     },
   },
-  "mcp": {
-    "exa": {
-      "type": "remote",
-      "url": "https://mcp.exa.ai/mcp",
-    },
-    "morph": {
-      "type": "local",
-      "command": ["bunx", "@morphllm/morphmcp"],
-      "environment": {
-        "ENABLED_TOOLS": "warp_grep",
-      },
-    },
-  },
+  "mcp": {},
 }
diff --git a/packages/opencode/src/agent/agent.ts b/packages/opencode/src/agent/agent.ts
index 94127e51ce..04f50b0316 100644
--- a/packages/opencode/src/agent/agent.ts
+++ b/packages/opencode/src/agent/agent.ts
@@ -157,6 +157,51 @@ export namespace Agent {
         mode: "primary",
         builtIn: true,
       },
+      summary: {
+        name: "summary",
+        mode: "subagent",
+        options: {},
+        builtIn: true,
+        permission: agentPermission,
+        prompt: `You are a title generator. You output ONLY a thread title. Nothing else.
+
+<task>
+Generate a brief title that would help the user find this conversation later.
+
+Follow all rules in <rules>
+Use the <examples> so you know what a good title looks like.
+Your output must be:
+- A single line
+- ≤50 characters
+- No explanations
+</task>
+
+<rules>
+- Focus on the main topic or question the user needs to retrieve
+- Use -ing verbs for actions (Debugging, Implementing, Analyzing)
+- Keep exact: technical terms, numbers, filenames, HTTP codes
+- Remove: the, this, my, a, an
+- Never assume tech stack
+- Never use tools
+- NEVER respond to questions, just generate a title for the conversation
+- The title should NEVER include "summarizing" or "generating" when generating a title
+- DO NOT SAY YOU CANNOT GENERATE A TITLE OR COMPLAIN ABOUT THE INPUT
+- Always output something meaningful, even if the input is minimal.
+- If the user message is short or conversational (e.g. “hello”, “lol”, “whats up”, “hey”):
+  → create a title that reflects the user’s tone or intent (such as Greeting, Quick check-in, Light chat, Intro message, etc.)
+</rules>
+
+<examples>
+"hey" -> Greeting
+"debug 500 errors in production" → Debugging production 500 errors
+"refactor user service" → Refactoring user service
+"why is app.js failing" → Analyzing app.js failure
+"implement rate limiting" → Implementing rate limiting
+"how do I connect postgres to my API" → Connecting Postgres to API
+"best practices for React hooks" → React hooks best practices
+</examples>`,
+        tools: {},
+      },
       plan: {
         name: "plan",
         options: {},
diff --git a/packages/opencode/src/session/compaction.ts b/packages/opencode/src/session/compaction.ts
index 54fe37ed4e..de3c3dca32 100644
--- a/packages/opencode/src/session/compaction.ts
+++ b/packages/opencode/src/session/compaction.ts
@@ -129,7 +129,7 @@ export namespace SessionCompaction {
     })
     const agent = await Agent.get(input.agent)
     const result = await processor.process({
-      requestID: input.parentID,
+      user: input.messages.findLast((m) => m.info.id === input.parentID)!.info as MessageV2.User,
       agent,
       abort: input.abort,
       sessionID: input.sessionID,
diff --git a/packages/opencode/src/session/llm.ts b/packages/opencode/src/session/llm.ts
index cda310e3a9..8277f36b1a 100644
--- a/packages/opencode/src/session/llm.ts
+++ b/packages/opencode/src/session/llm.ts
@@ -8,6 +8,7 @@ import { Instance } from "@/project/instance"
 import type { Agent } from "@/agent/agent"
 import type { MessageV2 } from "./message-v2"
 import { Plugin } from "@/plugin"
+import { SystemPrompt } from "./system"
 
 export namespace LLM {
   const log = Log.create({ service: "llm" })
@@ -22,6 +23,7 @@ export namespace LLM {
     system: string[]
     abort: AbortSignal
     messages: ModelMessage[]
+    small?: boolean
     tools: Record<string, Tool>
     retries?: number
   }
@@ -29,9 +31,19 @@ export namespace LLM {
   export type StreamOutput = StreamTextResult<ToolSet, unknown>
 
   export async function stream(input: StreamInput) {
+    const l = log
+      .clone()
+      .tag("providerID", input.model.providerID)
+      .tag("modelID", input.model.id)
+      .tag("sessionID", input.sessionID)
+      .tag("small", (input.small ?? false).toString())
+    l.info("stream", {
+      modelID: input.model.id,
+      providerID: input.model.providerID,
+    })
     const [language, cfg] = await Promise.all([Provider.getLanguage(input.model), Config.get()])
 
-    const [first, ...rest] = input.system
+    const [first, ...rest] = [...SystemPrompt.header(input.model.providerID), ...input.system]
     const system = [first, rest.join("\n")]
 
     const params = await Plugin.trigger(
@@ -49,13 +61,18 @@ export namespace LLM {
           : undefined,
         topP: input.agent.topP ?? ProviderTransform.topP(input.model),
         options: pipe(
-          ProviderTransform.options(input.model, input.sessionID),
+          mergeDeep(ProviderTransform.options(input.model, input.sessionID)),
+          input.small ? mergeDeep(ProviderTransform.smallOptions(input.model)) : mergeDeep({}),
           mergeDeep(input.model.options),
           mergeDeep(input.agent.options),
         ),
       },
     )
 
+    l.info("params", {
+      params,
+    })
+
     const maxOutputTokens = ProviderTransform.maxOutputTokens(
       input.model.api.npm,
       params.options,
@@ -65,14 +82,14 @@ export namespace LLM {
 
     return streamText({
       onError(error) {
-        log.error("stream error", {
+        l.error("stream error", {
           error,
         })
       },
       async experimental_repairToolCall(failed) {
         const lower = failed.toolCall.toolName.toLowerCase()
         if (lower !== failed.toolCall.toolName && input.tools[lower]) {
-          log.info("repairing tool call", {
+          l.info("repairing tool call", {
             tool: failed.toolCall.toolName,
             repaired: lower,
           })
@@ -94,6 +111,7 @@ export namespace LLM {
       topP: params.topP,
       providerOptions: ProviderTransform.providerOptions(input.model, params.options, input.messages),
       activeTools: Object.keys(input.tools).filter((x) => x !== "invalid"),
+      tools: input.tools,
       maxOutputTokens,
       abortSignal: input.abort,
       headers: {
diff --git a/packages/opencode/src/session/prompt.ts b/packages/opencode/src/session/prompt.ts
index 96a0435deb..8c4f4ecbad 100644
--- a/packages/opencode/src/session/prompt.ts
+++ b/packages/opencode/src/session/prompt.ts
@@ -41,6 +41,8 @@ import { fn } from "@/util/fn"
 import { SessionProcessor } from "./processor"
 import { TaskTool } from "@/tool/task"
 import { SessionStatus } from "./status"
+import { LLM } from "./llm"
+import { iife } from "@/util/iife"
 
 // @ts-ignore
 globalThis.AI_SDK_LOG_WARNINGS = false
@@ -281,7 +283,6 @@ export namespace SessionPrompt {
         })
 
       const model = await Provider.getModel(lastUser.model.providerID, lastUser.model.modelID)
-      const language = await Provider.getLanguage(model)
       const task = tasks.pop()
 
       // pending subtask
@@ -427,7 +428,6 @@ export namespace SessionPrompt {
       }
 
       // normal processing
-      const cfg = await Config.get()
       const agent = await Agent.get(lastUser.agent)
       const maxSteps = agent.maxSteps ?? Infinity
       const isLastStep = step >= maxSteps
@@ -435,6 +435,7 @@ export namespace SessionPrompt {
         messages: msgs,
         agent,
       })
+
       const processor = SessionProcessor.create({
         assistantMessage: (await Session.updateMessage({
           id: Identifier.ascending("message"),
@@ -467,7 +468,6 @@ export namespace SessionPrompt {
         model,
         agent,
         system: lastUser.system,
-        isLastStep,
       })
       const tools = await resolveTools({
         agent,
@@ -526,13 +526,9 @@ export namespace SessionPrompt {
     return Provider.defaultModel()
   }
 
-  async function resolveSystemPrompt(input: {
-    system?: string
-    agent: Agent.Info
-    model: Provider.Model
-    isLastStep?: boolean
-  }) {
-    let system = SystemPrompt.header(input.model.providerID)
+  async function resolveSystemPrompt(input: { system?: string; agent: Agent.Info; model: Provider.Model }) {
+    using _ = log.time("system")
+    let system = []
     system.push(
       ...(() => {
         if (input.system) return [input.system]
@@ -542,14 +538,6 @@ export namespace SessionPrompt {
     )
     system.push(...(await SystemPrompt.environment()))
     system.push(...(await SystemPrompt.custom()))
-
-    if (input.isLastStep) {
-      system.push(MAX_STEPS)
-    }
-
-    // max 2 system prompt messages for caching purposes
-    const [first, ...rest] = system
-    system = [first, rest.join("\n")]
     return system
   }
 
@@ -560,6 +548,7 @@ export namespace SessionPrompt {
     tools?: Record<string, boolean>
     processor: SessionProcessor.Info
   }) {
+    using _ = log.time("resolveTools")
     const tools: Record<string, AITool> = {}
     const enabledTools = pipe(
       input.agent.tools,
@@ -1319,28 +1308,24 @@ export namespace SessionPrompt {
       input.history.filter((m) => m.info.role === "user" && !m.parts.every((p) => "synthetic" in p && p.synthetic))
         .length === 1
     if (!isFirst) return
-    const cfg = await Config.get()
-    const small =
-      (await Provider.getSmallModel(input.providerID)) ?? (await Provider.getModel(input.providerID, input.modelID))
-    const language = await Provider.getLanguage(small)
-    const provider = await Provider.getProvider(small.providerID)
-    const options = pipe(
-      {},
-      mergeDeep(ProviderTransform.options(small, input.session.id, provider?.options)),
-      mergeDeep(ProviderTransform.smallOptions(small)),
-      mergeDeep(small.options),
-    )
-    await generateText({
-      // use higher # for reasoning models since reasoning tokens eat up a lot of the budget
-      maxOutputTokens: small.capabilities.reasoning ? 3000 : 20,
-      providerOptions: ProviderTransform.providerOptions(small, options, []),
+    const agent = await Agent.get("summary")
+    if (!agent) return
+    const result = await LLM.stream({
+      agent,
+      user: input.message.info as MessageV2.User,
+      system: [agent.prompt!],
+      small: true,
+      tools: {},
+      model: await iife(async () => {
+        if (agent.model) return await Provider.getModel(agent.model.providerID, agent.model.modelID)
+        return (
+          (await Provider.getSmallModel(input.providerID)) ?? (await Provider.getModel(input.providerID, input.modelID))
+        )
+      }),
+      abort: new AbortController().signal,
+      sessionID: input.session.id,
+      retries: 2,
       messages: [
-        ...SystemPrompt.title(small.providerID).map(
-          (x): ModelMessage => ({
-            role: "system",
-            content: x,
-          }),
-        ),
         {
           role: "user",
           content: "Generate a title for this conversation:\n",
@@ -1364,32 +1349,19 @@ export namespace SessionPrompt {
           },
         ]),
       ],
-      headers: small.headers,
-      model: language,
-      experimental_telemetry: {
-        isEnabled: cfg.experimental?.openTelemetry,
-        metadata: {
-          userId: cfg.username ?? "unknown",
-          sessionId: input.session.id,
-        },
-      },
     })
-      .then((result) => {
-        if (result.text)
-          return Session.update(input.session.id, (draft) => {
-            const cleaned = result.text
-              .replace(/<think>[\s\S]*?<\/think>\s*/g, "")
-              .split("\n")
-              .map((line) => line.trim())
-              .find((line) => line.length > 0)
-            if (!cleaned) return
+    const text = await result.text.catch((err) => log.error("failed to generate title", { error: err }))
+    if (text)
+      return Session.update(input.session.id, (draft) => {
+        const cleaned = text
+          .replace(/<think>[\s\S]*?<\/think>\s*/g, "")
+          .split("\n")
+          .map((line) => line.trim())
+          .find((line) => line.length > 0)
+        if (!cleaned) return
 
-            const title = cleaned.length > 100 ? cleaned.substring(0, 97) + "..." : cleaned
-            draft.title = title
-          })
-      })
-      .catch((error) => {
-        log.error("failed to generate title", { error, model: small.id })
+        const title = cleaned.length > 100 ? cleaned.substring(0, 97) + "..." : cleaned
+        draft.title = title
       })
   }
 }
diff --git a/packages/opencode/src/tool/bash.ts b/packages/opencode/src/tool/bash.ts
index 6b0b9d4104..cd04814bfe 100644
--- a/packages/opencode/src/tool/bash.ts
+++ b/packages/opencode/src/tool/bash.ts
@@ -50,35 +50,36 @@ const parser = lazy(async () => {
   return p
 })
 
+const getShell = lazy(() => {
+  const s = process.env.SHELL
+  if (s) {
+    const basename = path.basename(s)
+    if (!new Set(["fish", "nu"]).has(basename)) {
+      return s
+    }
+  }
+
+  if (process.platform === "darwin") {
+    return "/bin/zsh"
+  }
+
+  if (process.platform === "win32") {
+    // Let Bun / Node pick COMSPEC (usually cmd.exe)
+    // or explicitly:
+    return process.env.COMSPEC || true
+  }
+
+  const bash = Bun.which("bash")
+  if (bash) {
+    return bash
+  }
+
+  return true
+})
+
 // TODO: we may wanna rename this tool so it works better on other shells
-
 export const BashTool = Tool.define("bash", async () => {
-  const shell = iife(() => {
-    const s = process.env.SHELL
-    if (s) {
-      const basename = path.basename(s)
-      if (!new Set(["fish", "nu"]).has(basename)) {
-        return s
-      }
-    }
-
-    if (process.platform === "darwin") {
-      return "/bin/zsh"
-    }
-
-    if (process.platform === "win32") {
-      // Let Bun / Node pick COMSPEC (usually cmd.exe)
-      // or explicitly:
-      return process.env.COMSPEC || true
-    }
-
-    const bash = Bun.which("bash")
-    if (bash) {
-      return bash
-    }
-
-    return true
-  })
+  const shell = getShell()
   log.info("bash tool using shell", { shell })
 
   return {
diff --git a/packages/opencode/src/tool/registry.ts b/packages/opencode/src/tool/registry.ts
index 7e440a78aa..647c742671 100644
--- a/packages/opencode/src/tool/registry.ts
+++ b/packages/opencode/src/tool/registry.ts
@@ -21,8 +21,11 @@ import { Plugin } from "../plugin"
 import { WebSearchTool } from "./websearch"
 import { CodeSearchTool } from "./codesearch"
 import { Flag } from "@/flag/flag"
+import { Log } from "@/util/log"
 
 export namespace ToolRegistry {
+  const log = Log.create({ service: "tool.registry" })
+
   export const state = Instance.state(async () => {
     const custom = [] as Tool.Info[]
     const glob = new Bun.Glob("tool/*.{js,ts}")
@@ -119,10 +122,13 @@ export namespace ToolRegistry {
           }
           return true
         })
-        .map(async (t) => ({
-          id: t.id,
-          ...(await t.init()),
-        })),
+        .map(async (t) => {
+          using _ = log.time(t.id)
+          return {
+            id: t.id,
+            ...(await t.init()),
+          }
+        }),
     )
     return result
   }