惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

腾讯CDC
N
Netflix TechBlog - Medium
aimingoo的专栏
aimingoo的专栏
P
Proofpoint News Feed
F
Fortinet All Blogs
大猫的无限游戏
大猫的无限游戏
I
InfoQ
V
V2EX
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
有赞技术团队
有赞技术团队
G
Google Developers Blog
L
LangChain Blog
博客园_首页
M
MIT News - Artificial intelligence
H
Hackread – Cybersecurity News, Data Breaches, AI and More
月光博客
月光博客
IT之家
IT之家
量子位
宝玉的分享
宝玉的分享
S
SegmentFault 最新的问题
Stack Overflow Blog
Stack Overflow Blog
V
Visual Studio Blog
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
雷峰网
雷峰网

Recent Commits to openclaw:main

test: merge chat side-result checks · openclaw/openclaw@ddd2c2a test: merge cron history checks · openclaw/openclaw@f7eb746 test: merge responsive navigation shell checks · openclaw/openclaw@c2e4b47 docs(changelog): add codex oauth fixes · openclaw/openclaw@628e6cd test: merge navigation routing cases · openclaw/openclaw@5d8cecb Tests: mock channel registry bundled fallback · openclaw/openclaw@2b08233 Secrets: avoid broad web search discovery for single plugin config · openclaw/openclaw@a464f59 test: merge config view browser checks · openclaw/openclaw@20cf511 fix(status): align oauth health with runtime · openclaw/openclaw@eed7116 feat: add macOS screen snapshots for monitor preview (#67954) thanks … · openclaw/openclaw@f377db1 fix: report shared auth scopes in hello-ok (#67810) thanks @BunsDev · openclaw/openclaw@0b6c39b Auto-reply: avoid eager bundled route fallback · openclaw/openclaw@3ea1bf4 Tests: narrow session binding contract setup · openclaw/openclaw@54e4e16 fix(macOS): enable undo/redo in webchat composer text input (#34962) · openclaw/openclaw@00951dc Tests: speed up channel setup promotion · openclaw/openclaw@82b529a Docs: refresh agent instructions · openclaw/openclaw@5775fe2 fix(auth): serialize OAuth refresh across agents to fix #26322 (#67876) · openclaw/openclaw@8e79080 test: allow ollama public surface boundary test · openclaw/openclaw@7d4f1a6 Docs: add test performance guardrails · openclaw/openclaw@89706d3 Tests: restore context-engine usage proof · openclaw/openclaw@e4c4f95 Tests: slim context engine runtime coverage · openclaw/openclaw@74c198f ci: retry failed custom checkouts · openclaw/openclaw@0ee5baf test: trim duplicate provider auth onboarding cases · openclaw/openclaw@1ffc02e matrix: fix sessions_spawn --thread subagent session spawning (#67643) · openclaw/openclaw@1ce2596 test: reduce auth choice fixture churn · openclaw/openclaw@857b9cd test: mock health status config boundaries · openclaw/openclaw@9d5ab4a test: mock onboard config io boundary · openclaw/openclaw@299694d test: mock legacy state plugin boundaries · openclaw/openclaw@2713089 test: mock channel install boundaries · openclaw/openclaw@b945248 test: mock doctor preview channel boundaries · openclaw/openclaw@b1a3ad4
docs: add parallel specialist lane guidance · openclaw/op...
jarvisdoes · 2026-05-02 · via Recent Commits to openclaw:main

@@ -0,0 +1,127 @@

1+

---

2+

summary: "Run parallel specialist agents without clogging shared model and tool capacity"

3+

title: "Parallel specialist lanes"

4+

sidebarTitle: "Specialist lanes"

5+

read_when:

6+

- You route group chats to dedicated agents

7+

- You want parallel work without one long task blocking every chat

8+

- You are designing a multi-agent operations setup

9+

status: active

10+

---

11+12+

Parallel specialist lanes let one Gateway route different chats or rooms to

13+

different agents, while keeping the user experience fast. The trick is to treat

14+

parallelism as a scarce-resource design problem, not just as "more agents".

15+16+

## First principles

17+18+

A specialist lane only improves throughput when it reduces contention for the

19+

real bottlenecks:

20+21+

- **Session locks**: only one run should mutate a given session at a time.

22+

- **Global model capacity**: all visible chat runs still share provider limits.

23+

- **Tool capacity**: shell, browser, network, and repository work can be slower

24+

than the model turn itself.

25+

- **Context budget**: long transcripts make every future turn slower and less

26+

focused.

27+

- **Ownership ambiguity**: duplicate agents doing the same job waste capacity.

28+29+

OpenClaw already serializes runs per session and caps global parallelism through

30+

the [command queue](/concepts/queue). Specialist lanes add policy on top:

31+

which agent owns which work, what stays in chat, and what becomes background

32+

work.

33+34+

## Recommended rollout

35+36+

### Phase 1: lane contracts + background heavy work

37+38+

Give every lane a written contract in its workspace and system prompt:

39+40+

- **Purpose**: the work this lane owns.

41+

- **Non-goals**: work it should hand off instead of attempting.

42+

- **Chat budget**: quick answers stay in chat; long tasks should acknowledge

43+

briefly, then run in a background sub-agent or task.

44+

- **Handoff rule**: when another lane owns the work, say where it should go and

45+

provide a compact handoff summary.

46+

- **Tool-risk rule**: prefer the smallest tool surface that can do the job.

47+48+

This is the cheapest phase and fixes most clogging: one coding job no longer

49+

turns the research lane into molasses, and each chat keeps its own context clean.

50+51+

### Phase 2: priority and concurrency controls

52+53+

Tune queue and model capacity around the business value of each lane:

54+55+

```json5

56+

{

57+

agents: {

58+

defaults: {

59+

maxConcurrent: 4,

60+

subagents: { maxConcurrent: 8 },

61+

},

62+

},

63+

messages: {

64+

queue: {

65+

mode: "collect",

66+

debounceMs: 1000,

67+

cap: 20,

68+

drop: "summarize",

69+

},

70+

},

71+

}

72+

```

73+74+

Use direct/personal chats and production-ops agents for high-priority work. Let

75+

research, drafting, and batch coding move to background tasks when the system is

76+

busy.

77+78+

### Phase 3: coordinator / traffic controller

79+80+

Add a small coordinator pattern once multiple lanes are active:

81+82+

- Track active lane tasks and owners.

83+

- Detect duplicate requests across groups.

84+

- Route handoff summaries between lanes.

85+

- Surface only blockers, completed results, and decisions the human must make.

86+87+

Do not start here. A coordinator without lane contracts just coordinates chaos.

88+89+

## Minimal lane contract template

90+91+

```md

92+

# Lane contract

93+94+

## Owns

95+96+

- <job this lane is responsible for>

97+98+

## Does not own

99+100+

- <work to hand off>

101+102+

## Chat budget

103+104+

- Answer quick questions directly.

105+

- For multi-step, slow, or tool-heavy work: acknowledge briefly, spawn/background

106+

the work, then return the result when complete.

107+108+

## Handoff

109+110+

If another lane owns the request, reply with:

111+112+

- target lane

113+

- objective

114+

- relevant context

115+

- exact next action

116+117+

## Tool posture

118+119+

Use the smallest tool surface that can complete the task. Avoid broad shell or

120+

network work unless this lane explicitly owns it.

121+

```

122+123+

## Related

124+125+

- [Multi-agent routing](/concepts/multi-agent)

126+

- [Command queue](/concepts/queue)

127+

- [Sub-agents](/tools/subagents)