feat(agent): auto-enable GLM-4.7 thinking mode
Add automatic thinking mode support for Z.AI GLM-4.x models:
- GLM-4.7: Preserved thinking (clear_thinking: false)
- GLM-4.5/4.6: Interleaved thinking (clear_thinking: true)
Uses Z.AI Cloud API format: thinking: { type: "enabled", clear_thinking: boolean }
Includes patches for pi-ai, pi-agent-core, and pi-coding-agent to pass
extraParams through the stream pipeline. User can override via config
or disable via --thinking off.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
mneves75
committed by
Peter Steinberger
Peter Steinberger
parent
3f93781b4b
commit
f7b32195cb
446
src/agents/pi-embedded-runner-extraparams.test.ts
Normal file
446
src/agents/pi-embedded-runner-extraparams.test.ts
Normal file
@@ -0,0 +1,446 @@
|
||||
import { describe, expect, it } from "vitest";
|
||||
import { resolveExtraParams } from "./pi-embedded-runner.js";
|
||||
|
||||
/**
|
||||
* Tests for resolveExtraParams - the function that auto-enables GLM-4.x thinking mode.
|
||||
*
|
||||
* Z.AI Cloud API format: thinking: { type: "enabled", clear_thinking: boolean }
|
||||
* - GLM-4.7: Preserved thinking (clear_thinking: false) - reasoning kept across turns
|
||||
* - GLM-4.5/4.6: Interleaved thinking (clear_thinking: true) - reasoning cleared each turn
|
||||
*
|
||||
* @see https://docs.z.ai/guides/capabilities/thinking-mode
|
||||
*/
|
||||
|
||||
describe("resolveExtraParams", () => {
|
||||
describe("GLM-4.7 preserved thinking (clear_thinking: false)", () => {
|
||||
it("auto-enables preserved thinking for zai/glm-4.7 with no config", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false, // Preserved thinking for GLM-4.7
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("auto-enables preserved thinking for zai/GLM-4.7 (case insensitive)", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "GLM-4.7",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe("GLM-4.5/4.6 interleaved thinking (clear_thinking: true)", () => {
|
||||
it("auto-enables interleaved thinking for zai/glm-4.5", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.5",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: true, // Interleaved thinking for GLM-4.5
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("auto-enables interleaved thinking for zai/glm-4.6", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.6",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: true, // Interleaved thinking for GLM-4.6
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("auto-enables interleaved thinking for zai/glm-4-flash", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4-flash",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: true, // Non-4.7 gets interleaved
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("auto-enables interleaved thinking for zai/glm-4.5-air", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.5-air",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: true,
|
||||
},
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe("config overrides", () => {
|
||||
it("respects explicit thinking config from user (disable thinking)", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {
|
||||
agent: {
|
||||
models: {
|
||||
"zai/glm-4.7": {
|
||||
params: {
|
||||
thinking: {
|
||||
type: "disabled",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "disabled",
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("preserves other params while adding thinking config", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {
|
||||
agent: {
|
||||
models: {
|
||||
"zai/glm-4.7": {
|
||||
params: {
|
||||
temperature: 0.7,
|
||||
max_tokens: 4096,
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
temperature: 0.7,
|
||||
max_tokens: 4096,
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("does not override explicit thinking config even if partial", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {
|
||||
agent: {
|
||||
models: {
|
||||
"zai/glm-4.7": {
|
||||
params: {
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
// User explicitly omitted clear_thinking
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
// Should use user's config exactly, not merge defaults
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
},
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe("non-GLM models", () => {
|
||||
it("returns undefined for anthropic/claude with no config", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "anthropic",
|
||||
modelId: "claude-3-opus",
|
||||
});
|
||||
|
||||
expect(result).toBeUndefined();
|
||||
});
|
||||
|
||||
it("returns undefined for openai/gpt-4 with no config", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "openai",
|
||||
modelId: "gpt-4",
|
||||
});
|
||||
|
||||
expect(result).toBeUndefined();
|
||||
});
|
||||
|
||||
it("passes through params for non-GLM models without modification", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {
|
||||
agent: {
|
||||
models: {
|
||||
"openai/gpt-4": {
|
||||
params: {
|
||||
logprobs: true,
|
||||
top_logprobs: 5,
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
provider: "openai",
|
||||
modelId: "gpt-4",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
logprobs: true,
|
||||
top_logprobs: 5,
|
||||
});
|
||||
});
|
||||
|
||||
it("does not auto-enable thinking for non-zai provider even with glm-4 model id", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "openai",
|
||||
modelId: "glm-4.7", // Even if model ID contains glm-4
|
||||
});
|
||||
|
||||
expect(result).toBeUndefined();
|
||||
});
|
||||
});
|
||||
|
||||
describe("edge cases", () => {
|
||||
it("handles empty config gracefully", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {},
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("handles config with empty models gracefully", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: { agent: { models: {} } },
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("model alias lookup uses exact provider/model key", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {
|
||||
agent: {
|
||||
models: {
|
||||
"zai/glm-4.7": {
|
||||
alias: "smart",
|
||||
params: {
|
||||
custom_param: "value",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
custom_param: "value",
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("treats thinking: null as explicit config (no auto-enable)", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {
|
||||
agent: {
|
||||
models: {
|
||||
"zai/glm-4.7": {
|
||||
params: {
|
||||
thinking: null,
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
});
|
||||
|
||||
// null is !== undefined, so we respect the explicit null config
|
||||
expect(result).toEqual({
|
||||
thinking: null,
|
||||
});
|
||||
});
|
||||
|
||||
it("handles GLM-4.7 variants (glm-4.7-flash, glm-4.7-plus)", () => {
|
||||
// GLM-4.7-flash should get preserved thinking (contains "glm-4.7")
|
||||
const flashResult = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7-flash",
|
||||
});
|
||||
|
||||
expect(flashResult).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false, // Preserved thinking for GLM-4.7 variants
|
||||
},
|
||||
});
|
||||
|
||||
// GLM-4.7-plus should also get preserved thinking
|
||||
const plusResult = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7-plus",
|
||||
});
|
||||
|
||||
expect(plusResult).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe("thinkLevel parameter", () => {
|
||||
it("thinkLevel: 'off' disables auto-enable for GLM-4.x", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
thinkLevel: "off",
|
||||
});
|
||||
|
||||
// Should NOT auto-enable thinking when user explicitly disabled it
|
||||
expect(result).toBeUndefined();
|
||||
});
|
||||
|
||||
it("thinkLevel: 'off' still passes through explicit config", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: {
|
||||
agent: {
|
||||
models: {
|
||||
"zai/glm-4.7": {
|
||||
params: {
|
||||
custom_param: "value",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
thinkLevel: "off",
|
||||
});
|
||||
|
||||
// Should pass through config params but NOT auto-add thinking
|
||||
expect(result).toEqual({
|
||||
custom_param: "value",
|
||||
});
|
||||
});
|
||||
|
||||
it("thinkLevel: 'low' allows auto-enable", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
thinkLevel: "low",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("thinkLevel: 'high' allows auto-enable", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.5",
|
||||
thinkLevel: "high",
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: true,
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
it("thinkLevel: undefined (not specified) allows auto-enable", () => {
|
||||
const result = resolveExtraParams({
|
||||
cfg: undefined,
|
||||
provider: "zai",
|
||||
modelId: "glm-4.7",
|
||||
// thinkLevel not specified
|
||||
});
|
||||
|
||||
expect(result).toEqual({
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: false,
|
||||
},
|
||||
});
|
||||
});
|
||||
});
|
||||
});
|
||||
@@ -93,6 +93,70 @@ import { normalizeUsage, type UsageLike } from "./usage.js";
|
||||
import { loadWorkspaceBootstrapFiles } from "./workspace.js";
|
||||
|
||||
// Optional features can be implemented as Pi extensions that run in the same Node process.
|
||||
|
||||
/**
|
||||
* Resolve provider-specific extraParams from model config.
|
||||
* Auto-enables thinking mode for GLM-4.x models unless explicitly disabled.
|
||||
*
|
||||
* For ZAI GLM-4.x models, we auto-enable thinking via the Z.AI Cloud API format:
|
||||
* thinking: { type: "enabled", clear_thinking: boolean }
|
||||
*
|
||||
* - GLM-4.7: Preserved thinking (clear_thinking: false) - reasoning kept across turns
|
||||
* - GLM-4.5/4.6: Interleaved thinking (clear_thinking: true) - reasoning cleared each turn
|
||||
*
|
||||
* Users can override via config:
|
||||
* agent.models["zai/glm-4.7"].params.thinking = { type: "disabled" }
|
||||
*
|
||||
* Or disable via runtime flag: --thinking off
|
||||
*
|
||||
* @see https://docs.z.ai/guides/capabilities/thinking-mode
|
||||
* @internal Exported for testing only
|
||||
*/
|
||||
export function resolveExtraParams(params: {
|
||||
cfg: ClawdbotConfig | undefined;
|
||||
provider: string;
|
||||
modelId: string;
|
||||
thinkLevel?: string;
|
||||
}): Record<string, unknown> | undefined {
|
||||
const modelKey = `${params.provider}/${params.modelId}`;
|
||||
const modelConfig = params.cfg?.agent?.models?.[modelKey];
|
||||
let extraParams = modelConfig?.params ? { ...modelConfig.params } : undefined;
|
||||
|
||||
// Auto-enable thinking for ZAI GLM-4.x models when not explicitly configured
|
||||
// Skip if user explicitly disabled thinking via --thinking off
|
||||
if (params.provider === "zai" && params.thinkLevel !== "off") {
|
||||
const modelIdLower = params.modelId.toLowerCase();
|
||||
const isGlm4 = modelIdLower.includes("glm-4");
|
||||
|
||||
if (isGlm4) {
|
||||
// Check if user has explicitly configured thinking params
|
||||
const hasThinkingConfig = extraParams?.thinking !== undefined;
|
||||
|
||||
if (!hasThinkingConfig) {
|
||||
// GLM-4.7 supports preserved thinking (reasoning kept across turns)
|
||||
// GLM-4.5/4.6 use interleaved thinking (reasoning cleared each turn)
|
||||
// Z.AI Cloud API format: thinking: { type: "enabled", clear_thinking: boolean }
|
||||
const isGlm47 = modelIdLower.includes("glm-4.7");
|
||||
const clearThinking = !isGlm47;
|
||||
|
||||
extraParams = {
|
||||
...extraParams,
|
||||
thinking: {
|
||||
type: "enabled",
|
||||
clear_thinking: clearThinking,
|
||||
},
|
||||
};
|
||||
|
||||
log.debug(
|
||||
`auto-enabled thinking for ${modelKey}: type=enabled, clear_thinking=${clearThinking}`,
|
||||
);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return extraParams;
|
||||
}
|
||||
|
||||
// We configure context pruning per-session via a WeakMap registry keyed by the SessionManager instance.
|
||||
|
||||
function resolvePiExtensionPath(id: string): string {
|
||||
@@ -837,6 +901,13 @@ export async function compactEmbeddedPiSession(params: {
|
||||
sandboxEnabled: !!sandbox?.enabled,
|
||||
});
|
||||
|
||||
const extraParams = resolveExtraParams({
|
||||
cfg: params.config,
|
||||
provider,
|
||||
modelId,
|
||||
thinkLevel: params.thinkLevel,
|
||||
});
|
||||
|
||||
let session: Awaited<ReturnType<typeof createAgentSession>>["session"];
|
||||
({ session } = await createAgentSession({
|
||||
cwd: resolvedWorkspace,
|
||||
@@ -853,6 +924,7 @@ export async function compactEmbeddedPiSession(params: {
|
||||
skills: [],
|
||||
contextFiles: [],
|
||||
additionalExtensionPaths,
|
||||
extraParams,
|
||||
}));
|
||||
|
||||
try {
|
||||
@@ -1148,6 +1220,13 @@ export async function runEmbeddedPiAgent(params: {
|
||||
sandboxEnabled: !!sandbox?.enabled,
|
||||
});
|
||||
|
||||
const extraParams = resolveExtraParams({
|
||||
cfg: params.config,
|
||||
provider,
|
||||
modelId,
|
||||
thinkLevel,
|
||||
});
|
||||
|
||||
let session: Awaited<
|
||||
ReturnType<typeof createAgentSession>
|
||||
>["session"];
|
||||
@@ -1168,6 +1247,7 @@ export async function runEmbeddedPiAgent(params: {
|
||||
skills: [],
|
||||
contextFiles: [],
|
||||
additionalExtensionPaths,
|
||||
extraParams,
|
||||
}));
|
||||
|
||||
try {
|
||||
|
||||
@@ -982,6 +982,8 @@ export type AuthConfig = {
|
||||
|
||||
export type AgentModelEntryConfig = {
|
||||
alias?: string;
|
||||
/** Provider-specific API parameters (e.g., GLM-4.7 thinking mode). */
|
||||
params?: Record<string, unknown>;
|
||||
};
|
||||
|
||||
export type AgentModelListConfig = {
|
||||
|
||||
@@ -910,6 +910,8 @@ export const ClawdbotSchema = z.object({
|
||||
z.string(),
|
||||
z.object({
|
||||
alias: z.string().optional(),
|
||||
/** Provider-specific API parameters (e.g., GLM-4.7 thinking mode). */
|
||||
params: z.record(z.string(), z.unknown()).optional(),
|
||||
}),
|
||||
)
|
||||
.optional(),
|
||||
|
||||
Reference in New Issue
Block a user