TLDR;

[Erro ao processar: Authentication failed for provider 'symgateway': I]

22
00:00:52,119 --> 00:00:55,799
tasks and you get a live view of what

23
00:00:54,280 --> 00:00:57,160
your agents are doing now, what's

24
00:00:55,799 --> 00:00:59,439
running, what just finished, what

25
00:00:57,159 --> 00:01:01,799
failed, recent task details right there.

26
00:00:59,439 --> 00:01:03,199
Before this, running multiple agents

27
00:01:01,799 --> 00:01:05,599
meant guessing. You'd fire something

28
00:01:03,200 --> 00:01:07,840
off, assume it worked, and only find out

29
00:01:05,599 --> 00:01:09,879
when it broke when something downstream

30
00:01:07,840 --> 00:01:11,359
went wrong. Agencies running OpenClaw

31
00:01:09,879 --> 00:01:13,599
for client work, for example, content,

32
00:01:11,359 --> 00:01:15,560
outreach, research, reporting, had no

33
00:01:13,599 --> 00:01:16,919
clean way to monitor what was actually

34
00:01:15,560 --> 00:01:18,960
happening

35
00:01:16,920 --> 00:01:21,120
without digging into the logs. Now, you

36
00:01:18,959 --> 00:01:23,919
can see it all in one place, live. And

37
00:01:21,120 --> 00:01:26,480
that's a shift in how you manage agents.

38
00:01:23,920 --> 00:01:28,480
Less babysitting, more visibility, and

39
00:01:26,480 --> 00:01:30,040
for anyone running OpenClaw at scale,

40
00:01:28,480 --> 00:01:33,359
that's exactly what was missing.

41
00:01:30,040 --> 00:01:36,080
Failover that actually fails over. So,

42
00:01:33,359 --> 00:01:38,400
this is another update from 4.1.

43
00:01:36,079 --> 00:01:40,679
OpenClaw has always had model failover,

44
00:01:38,400 --> 00:01:42,840
the ability to switch providers if one

45
00:01:40,680 --> 00:01:45,560
goes down or hits rate limits, but in

46
00:01:42,840 --> 00:01:48,359
practice, it had a big problem. So, when

47
00:01:45,560 --> 00:01:50,960
a provider hit a rate limit, OpenClaw

48
00:01:48,359 --> 00:01:53,840
would sometimes keep retrying the same

49
00:01:50,959 --> 00:01:56,280
off profile on the same provider over

50
00:01:53,840 --> 00:01:58,799
and over instead of just moving on. 4.1

51
00:01:56,280 --> 00:02:01,359
fixes that. It caps same provider

52
00:01:58,799 --> 00:02:03,000
retries before switching to a different

53
00:02:01,359 --> 00:02:06,159
model or provider. And there's a new

54
00:02:03,000 --> 00:02:08,879
setting, off.cooldowns.rate

55
00:02:06,159 --> 00:02:11,319
limited profile rotations. In English,

56
00:02:08,879 --> 00:02:14,400
that means so you can control exactly

57
00:02:11,319 --> 00:02:16,759
how aggressive that rotation is. What

58
00:02:14,400 --> 00:02:19,360
this means is basically in plain terms,

59
00:02:16,759 --> 00:02:21,399
your agents stop looping over. If one

60
00:02:19,360 --> 00:02:23,440
provider gets congested, OpenClaw

61
00:02:21,400 --> 00:02:25,439
switches cleanly, automatically, instead

62
00:02:23,439 --> 00:02:27,560
of hammering the same wall until

63
00:02:25,439 --> 00:02:30,039
everything times out. If you've ever had

64
00:02:27,560 --> 00:02:31,680
an OpenClaw workflow stall at like 2:00

65
00:02:30,039 --> 00:02:33,919
a.m. and then you wake up and your cron

66
00:02:31,680 --> 00:02:36,840
job didn't work properly, it's sometimes

67
00:02:33,919 --> 00:02:39,399
because a model hit capacity and this is

68
00:02:36,840 --> 00:02:41,159
a fix that may help you in the future.

69
00:02:39,400 --> 00:02:46,920
Now, there's also improvements on web

70
00:02:41,159 --> 00:02:51,199
search. So, 4.1 ships a bundled SearXNG

71
00:02:46,919 --> 00:02:53,719
plugin for web search. SearXNG

72
00:02:51,199 --> 00:02:55,879
is an open-source search engine where

73
00:02:53,719 --> 00:02:58,479
you can self-host it, right? Now, this

74
00:02:55,879 --> 00:03:01,759
basically means the agents can now

75
00:02:58,479 --> 00:03:03,560
search the web through a provider that

76
00:03:01,759 --> 00:03:05,799
you control. It's not just whatever

77
00:03:03,560 --> 00:03:08,400
search API you happen to bolt on. You

78
00:03:05,800 --> 00:03:10,960
can configure the host, point OpenClaw

79
00:03:08,400 --> 00:03:13,280
at it, and your agents have a clean,

80
00:03:10,960 --> 00:03:15,080
private, configurable web search built

81
00:03:13,280 --> 00:03:17,000
in. So, for example, if you're a solo

82
00:03:15,080 --> 00:03:19,160
operator and you're running agents to do

83
00:03:17,000 --> 00:03:21,439
competitive research or content ideation

84
00:03:19,159 --> 00:03:23,120
or market monitoring, this would help

85
00:03:21,439 --> 00:03:25,240
you because you're not dependent on

86
00:03:23,120 --> 00:03:27,000
third-party search APIs. You control the

87
00:03:25,240 --> 00:03:29,120
data, you control the cost, and that's a

88
00:03:27,000 --> 00:03:31,879
meaningful upgrade for anyone running

89
00:03:29,120 --> 00:03:34,879
OpenClaw in client-facing workflows.

90
00:03:31,879 --> 00:03:38,079
Now, if you or your clients are on AWS

91
00:03:34,879 --> 00:03:40,680
Bedrock, 4.1 also adds native guardrail

92
00:03:38,080 --> 00:03:43,400
support. So, Bedrock guardrails let you

93
00:03:40,680 --> 00:03:45,319
set content filters, topic restrictions,

94
00:03:43,400 --> 00:03:47,680
and safety rules at the infrastructure

95
00:03:45,319 --> 00:03:50,000
level before responses even reach your

96
00:03:47,680 --> 00:03:52,159
users. So, now OpenClaw plugs directly

97
00:03:50,000 --> 00:03:53,719
into all of that. For agencies running

98
00:03:52,159 --> 00:03:55,680
OpenClaw, for example, for enterprise

99
00:03:53,719 --> 00:03:57,400
clients, this is a compliance win

100
00:03:55,680 --> 00:04:00,480
because you can tell a client, "Okay,

101
00:03:57,400 --> 00:04:02,760
our AI agents run through AWS with

102
00:04:00,479 --> 00:04:05,119
guardrails active, content controls,

103
00:04:02,759 --> 00:04:06,959
audit trails, enterprise-grade safety,

104
00:04:05,120 --> 00:04:08,920
basically." That's a much easier

105
00:04:06,960 --> 00:04:11,320
conversation than trying to explain how

106
00:04:08,919 --> 00:04:12,879
you're manually reviewing agent outputs.

107
00:04:11,319 --> 00:04:15,319
So, better infrastructure for safety and

108
00:04:12,879 --> 00:04:17,199
security. Now, you can also say, "Hey,"

109
00:04:15,319 --> 00:04:18,879
and your agent wakes up. This is a small

110
00:04:17,199 --> 00:04:22,159
little feature, but it's worth knowing.

111
00:04:18,879 --> 00:04:25,040
4.1 adds voice wake on macOS. You can

112
00:04:22,160 --> 00:04:27,120
now trigger talk mode with a wake word

113
00:04:25,040 --> 00:04:29,680
hands-free. If you're working at your

114
00:04:27,120 --> 00:04:31,639
desk, screen sharing, or on a call and

115
00:04:29,680 --> 00:04:33,400
you want to fire a you can. It's a

116
00:04:31,639 --> 00:04:35,959
little small thing, but it's the kind of

117
00:04:33,399 --> 00:04:38,000
small thing that makes the tool feel

118
00:04:35,959 --> 00:04:39,479
more like a real assistant and less like

119
00:04:38,000 --> 00:04:41,800
a piece of software you have to babysit

120
00:04:39,480 --> 00:04:44,080
all the time. Cron jobs also have their

121
00:04:41,800 --> 00:04:45,720
own tool allow list now, and this one is

122
00:04:44,079 --> 00:04:49,639
for anyone running scheduled agent

123
00:04:45,720 --> 00:04:49,640
tasks. So, 4.1

124
00:04:49,959 --> 00:04:55,839
adds OpenClaw cron tools, which is

125
00:04:52,439 --> 00:04:58,639
basically a per job tool allow list for

126
00:04:55,839 --> 00:05:01,039
cron scheduled agent runs. What that

127
00:04:58,639 --> 00:05:03,599
means in English is when you schedule a

128
00:05:01,040 --> 00:05:05,640
recurring agent task, you can now lock

129
00:05:03,600 --> 00:05:08,160
it down to only the tools actually

130
00:05:05,639 --> 00:05:09,759
needs. So, a content agent that runs

131
00:05:08,160 --> 00:05:12,000
every morning to pull research and draft

132
00:05:09,759 --> 00:05:14,000
a post only needs web search and a file

133
00:05:12,000 --> 00:05:16,920
writer. It doesn't need like executive

134
00:05:14,000 --> 00:05:19,519
permissions on external API access. And

135
00:05:16,920 --> 00:05:21,680
now you can explain that explicitly to

136
00:05:19,519 --> 00:05:23,439
your agent per job, which means you get

137
00:05:21,680 --> 00:05:24,759
tighter control, a smaller attack

138
00:05:23,439 --> 00:05:27,519
surface as well. And if something goes

139
00:05:24,759 --> 00:05:29,279
wrong, it can only do it wrong inside

140
00:05:27,519 --> 00:05:30,759
the little box that you gave it. And

141
00:05:29,279 --> 00:05:32,599
also sometimes like you'll find these

142
00:05:30,759 --> 00:05:34,319
agents they try and call different APIs

143
00:05:32,600 --> 00:05:36,120
for no reason. It's like, "Stop doing

144
00:05:34,319 --> 00:05:37,759
that. Like it's inefficient." And they

145
00:05:36,120 --> 00:05:40,240
don't realize. So, if you can explain it

146
00:05:37,759 --> 00:05:42,639
to the agent like this with this update,

147
00:05:40,240 --> 00:05:46,720
then it just helps a lot. Now, also Z

148
00:05:42,639 --> 00:05:48,319
AI's GLM 5.1 and GLM 5V Turbo that

149
00:05:46,720 --> 00:05:51,600
literally just dropped are now in

150
00:05:48,319 --> 00:05:54,879
OpenClaw's bundled provider catalog. So,

151
00:05:51,600 --> 00:05:56,960
GLM 5.1 is a strong general model. It's

152
00:05:54,879 --> 00:06:00,399
fast, it's capable, it's good at

153
00:05:56,959 --> 00:06:02,919
structured output. GLM 5V Turbo is the

154
00:06:00,399 --> 00:06:05,000
vision variant. Both are available now

155
00:06:02,920 --> 00:06:06,720
through the Z AI provider. I think you

156
00:06:05,000 --> 00:06:07,959
can also use them on Open Router as

157
00:06:06,720 --> 00:06:10,400
well. And if you've been running

158
00:06:07,959 --> 00:06:12,319
OpenClaw with a limited model setup, so

159
00:06:10,399 --> 00:06:15,319
for example, maybe you used Claude or

160
00:06:12,319 --> 00:06:16,719
GPT or a llama, right? Adding GLM is a

161
00:06:15,319 --> 00:06:18,439
really powerful API. It's great as a

162
00:06:16,720 --> 00:06:20,440
sub-agent and you can add it to the

163
00:06:18,439 --> 00:06:22,240
rotation and it gives you another option

164
00:06:20,439 --> 00:06:24,079
for cost management. So, when I was

165
00:06:22,240 --> 00:06:26,519
talking about sub-agents, for example,

166
00:06:24,079 --> 00:06:28,879
you could run cheaper tasks to GLM

167
00:06:26,519 --> 00:06:30,319
sub-agents using that new API and save

168
00:06:28,879 --> 00:06:31,279
your heavier models for more complex

169
00:06:30,319 --> 00:06:33,399
work. So, you could have like, for

170
00:06:31,279 --> 00:06:35,879
example, Claude as the brain and then

171
00:06:33,399 --> 00:06:37,879
that funnels smaller sub-tasks to a

172
00:06:35,879 --> 00:06:39,800
cheaper API like GLM. And that's how

173
00:06:37,879 --> 00:06:41,360
multi-modal

174
00:06:39,800 --> 00:06:43,680
models and and setups should work,

175
00:06:41,360 --> 00:06:45,439
matching the task to the model, not just

176
00:06:43,680 --> 00:06:46,840
defaulting to one provider for

177
00:06:45,439 --> 00:06:48,680
everything because some of these are

178
00:06:46,839 --> 00:06:50,159
quite overpowered if that makes sense

179
00:06:48,680 --> 00:06:52,439
and expensive, too. And it's better to

180
00:06:50,160 --> 00:06:54,360
route your tasks to cheaper APIs that

181
00:06:52,439 --> 00:06:56,399
are way easier to use. And if you're

182
00:06:54,360 --> 00:06:59,000
trying to figure out how to actually set

183
00:06:56,399 --> 00:07:00,839
OpenClaw up, run it reliably, and use it

184
00:06:59,000 --> 00:07:02,319
for real business workflows, come join

185
00:07:00,839 --> 00:07:04,159
us in the Air Profit Bootcamp. We've got

186
00:07:02,319 --> 00:07:06,839
a 30-day roadmap built specifically

187
00:07:04,160 --> 00:07:09,080
around OpenClaw setup, agent design, how

188
00:07:06,839 --> 00:07:11,319
to use the {forward slash} tasks board,

189
00:07:09,079 --> 00:07:13,359
how to structure cron jobs and schedule

190
00:07:11,319 --> 00:07:15,439
tasks, how to set up a failover so your

191
00:07:13,360 --> 00:07:16,879
automations don't loop out at 3:00 a.m.,

192
00:07:15,439 --> 00:07:18,879
daily tutorials walking you through

193
00:07:16,879 --> 00:07:20,519
every update as it ships, and four

194
00:07:18,879 --> 00:07:22,800
coaching calls every week with people

195
00:07:20,519 --> 00:07:24,879
who are deep in OpenClaw right now.

196
00:07:22,800 --> 00:07:26,800
There's 2,700 business owners in there,

197
00:07:24,879 --> 00:07:28,839
a lot of them already running OpenClaw

198
00:07:26,800 --> 00:07:30,840
for their business or for client work,

199
00:07:28,839 --> 00:07:32,679
for lead gen, for content, for research.

200
00:07:30,839 --> 00:07:34,679
Link in the comments description or just

201
00:07:32,680 --> 00:07:37,360
go to the airprofitbootcamp.com to get

202
00:07:34,680 --> 00:07:39,000
access. So, there's also over 40 sick

203
00:07:37,360 --> 00:07:40,920
fixes like I was talking about, and

204
00:07:39,000 --> 00:07:43,639
here's what to actually pay attention

205
00:07:40,920 --> 00:07:46,439
to. Beyond the new features, 4.1 is a

206
00:07:43,639 --> 00:07:48,079
serious stability release, right? And

207
00:07:46,439 --> 00:07:51,120
let me flag the ones that matter most

208
00:07:48,079 --> 00:07:53,

Formato de saida (IMPORTANTE - siga exatamente):
Escreva APENAS as 3 respostas, sem as perguntas, no formato:
[resposta 1]. [resposta 2]. [resposta 3]

NAO inclua as perguntas na saida, apenas as respostas em italico separadas por ponto. NAO comece as respostas com SIM ou NÃO. Va direto ao ponto.
}

Resumo

[Erro ao processar: Authentication failed for provider 'symgateway': I]

22
00:00:52,119 --> 00:00:55,799
tasks and you get a live view of what

23
00:00:54,280 --> 00:00:57,160
your agents are doing now, what's

24
00:00:55,799 --> 00:00:59,439
running, what just finished, what

25
00:00:57,159 --> 00:01:01,799
failed, recent task details right there.

26
00:00:59,439 --> 00:01:03,199
Before this, running multiple agents

27
00:01:01,799 --> 00:01:05,599
meant guessing. You'd fire something

28
00:01:03,200 --> 00:01:07,840
off, assume it worked, and only find out

29
00:01:05,599 --> 00:01:09,879
when it broke when something downstream

30
00:01:07,840 --> 00:01:11,359
went wrong. Agencies running OpenClaw

31
00:01:09,879 --> 00:01:13,599
for client work, for example, content,

32
00:01:11,359 --> 00:01:15,560
outreach, research, reporting, had no

33
00:01:13,599 --> 00:01:16,919
clean way to monitor what was actually

34
00:01:15,560 --> 00:01:18,960
happening

35
00:01:16,920 --> 00:01:21,120
without digging into the logs. Now, you

36
00:01:18,959 --> 00:01:23,919
can see it all in one place, live. And

37
00:01:21,120 --> 00:01:26,480
that's a shift in how you manage agents.

38
00:01:23,920 --> 00:01:28,480
Less babysitting, more visibility, and

39
00:01:26,480 --> 00:01:30,040
for anyone running OpenClaw at scale,

40
00:01:28,480 --> 00:01:33,359
that's exactly what was missing.

41
00:01:30,040 --> 00:01:36,080
Failover that actually fails over. So,

42
00:01:33,359 --> 00:01:38,400
this is another update from 4.1.

43
00:01:36,079 --> 00:01:40,679
OpenClaw has always had model failover,

44
00:01:38,400 --> 00:01:42,840
the ability to switch providers if one

45
00:01:40,680 --> 00:01:45,560
goes down or hits rate limits, but in

46
00:01:42,840 --> 00:01:48,359
practice, it had a big problem. So, when

47
00:01:45,560 --> 00:01:50,960
a provider hit a rate limit, OpenClaw

48
00:01:48,359 --> 00:01:53,840
would sometimes keep retrying the same

49
00:01:50,959 --> 00:01:56,280
off profile on the same provider over

50
00:01:53,840 --> 00:01:58,799
and over instead of just moving on. 4.1

51
00:01:56,280 --> 00:02:01,359
fixes that. It caps same provider

52
00:01:58,799 --> 00:02:03,000
retries before switching to a different

53
00:02:01,359 --> 00:02:06,159
model or provider. And there's a new

54
00:02:03,000 --> 00:02:08,879
setting, off.cooldowns.rate

55
00:02:06,159 --> 00:02:11,319
limited profile rotations. In English,

56
00:02:08,879 --> 00:02:14,400
that means so you can control exactly

57
00:02:11,319 --> 00:02:16,759
how aggressive that rotation is. What

58
00:02:14,400 --> 00:02:19,360
this means is basically in plain terms,

59
00:02:16,759 --> 00:02:21,399
your agents stop looping over. If one

60
00:02:19,360 --> 00:02:23,440
provider gets congested, OpenClaw

61
00:02:21,400 --> 00:02:25,439
switches cleanly, automatically, instead

62
00:02:23,439 --> 00:02:27,560
of hammering the same wall until

63
00:02:25,439 --> 00:02:30,039
everything times out. If you've ever had

64
00:02:27,560 --> 00:02:31,680
an OpenClaw workflow stall at like 2:00

65
00:02:30,039 --> 00:02:33,919
a.m. and then you wake up and your cron

66
00:02:31,680 --> 00:02:36,840
job didn't work properly, it's sometimes

67
00:02:33,919 --> 00:02:39,399
because a model hit capacity and this is

68
00:02:36,840 --> 00:02:41,159
a fix that may help you in the future.

69
00:02:39,400 --> 00:02:46,920
Now, there's also improvements on web

70
00:02:41,159 --> 00:02:51,199
search. So, 4.1 ships a bundled SearXNG

71
00:02:46,919 --> 00:02:53,719
plugin for web search. SearXNG

72
00:02:51,199 --> 00:02:55,879
is an open-source search engine where

73
00:02:53,719 --> 00:02:58,479
you can self-host it, right? Now, this

74
00:02:55,879 --> 00:03:01,759
basically means the agents can now

75
00:02:58,479 --> 00:03:03,560
search the web through a provider that

76
00:03:01,759 --> 00:03:05,799
you control. It's not just whatever

77
00:03:03,560 --> 00:03:08,400
search API you happen to bolt on. You

78
00:03:05,800 --> 00:03:10,960
can configure the host, point OpenClaw

79
00:03:08,400 --> 00:03:13,280
at it, and your agents have a clean,

80
00:03:10,960 --> 00:03:15,080
private, configurable web search built

81
00:03:13,280 --> 00:03:17,000
in. So, for example, if you're a solo

82
00:03:15,080 --> 00:03:19,160
operator and you're running agents to do

83
00:03:17,000 --> 00:03:21,439
competitive research or content ideation

84
00:03:19,159 --> 00:03:23,120
or market monitoring, this would help

85
00:03:21,439 --> 00:03:25,240
you because you're not dependent on

86
00:03:23,120 --> 00:03:27,000
third-party search APIs. You control the

87
00:03:25,240 --> 00:03:29,120
data, you control the cost, and that's a

88
00:03:27,000 --> 00:03:31,879
meaningful upgrade for anyone running

89
00:03:29,120 --> 00:03:34,879
OpenClaw in client-facing workflows.

90
00:03:31,879 --> 00:03:38,079
Now, if you or your clients are on AWS

91
00:03:34,879 --> 00:03:40,680
Bedrock, 4.1 also adds native guardrail

92
00:03:38,080 --> 00:03:43,400
support. So, Bedrock guardrails let you

93
00:03:40,680 --> 00:03:45,319
set content filters, topic restrictions,

94
00:03:43,400 --> 00:03:47,680
and safety rules at the infrastructure

95
00:03:45,319 --> 00:03:50,000
level before responses even reach your

96
00:03:47,680 --> 00:03:52,159
users. So, now OpenClaw plugs directly

97
00:03:50,000 --> 00:03:53,719
into all of that. For agencies running

98
00:03:52,159 --> 00:03:55,680
OpenClaw, for example, for enterprise

99
00:03:53,719 --> 00:03:57,400
clients, this is a compliance win

100
00:03:55,680 --> 00:04:00,480
because you can tell a client, "Okay,

101
00:03:57,400 --> 00:04:02,760
our AI agents run through AWS with

102
00:04:00,479 --> 00:04:05,119
guardrails active, content controls,

103
00:04:02,759 --> 00:04:06,959
audit trails, enterprise-grade safety,

104
00:04:05,120 --> 00:04:08,920
basically." That's a much easier

105
00:04:06,960 --> 00:04:11,320
conversation than trying to explain how

106
00:04:08,919 --> 00:04:12,879
you're manually reviewing agent outputs.

107
00:04:11,319 --> 00:04:15,319
So, better infrastructure for safety and

108
00:04:12,879 --> 00:04:17,199
security. Now, you can also say, "Hey,"

109
00:04:15,319 --> 00:04:18,879
and your agent wakes up. This is a small

110
00:04:17,199 --> 00:04:22,159
little feature, but it's worth knowing.

111
00:04:18,879 --> 00:04:25,040
4.1 adds voice wake on macOS. You can

112
00:04:22,160 --> 00:04:27,120
now trigger talk mode with a wake word

113
00:04:25,040 --> 00:04:29,680
hands-free. If you're working at your

114
00:04:27,120 --> 00:04:31,639
desk, screen sharing, or on a call and

115
00:04:29,680 --> 00:04:33,400
you want to fire a you can. It's a

116
00:04:31,639 --> 00:04:35,959
little small thing, but it's the kind of

117
00:04:33,399 --> 00:04:38,000
small thing that makes the tool feel

118
00:04:35,959 --> 00:04:39,479
more like a real assistant and less like

119
00:04:38,000 --> 00:04:41,800
a piece of software you have to babysit

120
00:04:39,480 --> 00:04:44,080
all the time. Cron jobs also have their

121
00:04:41,800 --> 00:04:45,720
own tool allow list now, and this one is

122
00:04:44,079 --> 00:04:49,639
for anyone running scheduled agent

123
00:04:45,720 --> 00:04:49,640
tasks. So, 4.1

124
00:04:49,959 --> 00:04:55,839
adds OpenClaw cron tools, which is

125
00:04:52,439 --> 00:04:58,639
basically a per job tool allow list for

126
00:04:55,839 --> 00:05:01,039
cron scheduled agent runs. What that

127
00:04:58,639 --> 00:05:03,599
means in English is when you schedule a

128
00:05:01,040 --> 00:05:05,640
recurring agent task, you can now lock

129
00:05:03,600 --> 00:05:08,160
it down to only the tools actually

130
00:05:05,639 --> 00:05:09,759
needs. So, a content agent that runs

131
00:05:08,160 --> 00:05:12,000
every morning to pull research and draft

132
00:05:09,759 --> 00:05:14,000
a post only needs web search and a file

133
00:05:12,000 --> 00:05:16,920
writer. It doesn't need like executive

134
00:05:14,000 --> 00:05:19,519
permissions on external API access. And

135
00:05:16,920 --> 00:05:21,680
now you can explain that explicitly to

136
00:05:19,519 --> 00:05:23,439
your agent per job, which means you get

137
00:05:21,680 --> 00:05:24,759
tighter control, a smaller attack

138
00:05:23,439 --> 00:05:27,519
surface as well. And if something goes

139
00:05:24,759 --> 00:05:29,279
wrong, it can only do it wrong inside

140
00:05:27,519 --> 00:05:30,759
the little box that you gave it. And

141
00:05:29,279 --> 00:05:32,599
also sometimes like you'll find these

142
00:05:30,759 --> 00:05:34,319
agents they try and call different APIs

143
00:05:32,600 --> 00:05:36,120
for no reason. It's like, "Stop doing

144
00:05:34,319 --> 00:05:37,759
that. Like it's inefficient." And they

145
00:05:36,120 --> 00:05:40,240
don't realize. So, if you can explain it

146
00:05:37,759 --> 00:05:42,639
to the agent like this with this update,

147
00:05:40,240 --> 00:05:46,720
then it just helps a lot. Now, also Z

148
00:05:42,639 --> 00:05:48,319
AI's GLM 5.1 and GLM 5V Turbo that

149
00:05:46,720 --> 00:05:51,600
literally just dropped are now in

150
00:05:48,319 --> 00:05:54,879
OpenClaw's bundled provider catalog. So,

151
00:05:51,600 --> 00:05:56,960
GLM 5.1 is a strong general model. It's

152
00:05:54,879 --> 00:06:00,399
fast, it's capable, it's good at

153
00:05:56,959 --> 00:06:02,919
structured output. GLM 5V Turbo is the

154
00:06:00,399 --> 00:06:05,000
vision variant. Both are available now

155
00:06:02,920 --> 00:06:06,720
through the Z AI provider. I think you

156
00:06:05,000 --> 00:06:07,959
can also use them on Open Router as

157
00:06:06,720 --> 00:06:10,400
well. And if you've been running

158
00:06:07,959 --> 00:06:12,319
OpenClaw with a limited model setup, so

159
00:06:10,399 --> 00:06:15,319
for example, maybe you used Claude or

160
00:06:12,319 --> 00:06:16,719
GPT or a llama, right? Adding GLM is a

161
00:06:15,319 --> 00:06:18,439
really powerful API. It's great as a

162
00:06:16,720 --> 00:06:20,440
sub-agent and you can add it to the

163
00:06:18,439 --> 00:06:22,240
rotation and it gives you another option

164
00:06:20,439 --> 00:06:24,079
for cost management. So, when I was

165
00:06:22,240 --> 00:06:26,519
talking about sub-agents, for example,

166
00:06:24,079 --> 00:06:28,879
you could run cheaper tasks to GLM

167
00:06:26,519 --> 00:06:30,319
sub-agents using that new API and save

168
00:06:28,879 --> 00:06:31,279
your heavier models for more complex

169
00:06:30,319 --> 00:06:33,399
work. So, you could have like, for

170
00:06:31,279 --> 00:06:35,879
example, Claude as the brain and then

171
00:06:33,399 --> 00:06:37,879
that funnels smaller sub-tasks to a

172
00:06:35,879 --> 00:06:39,800
cheaper API like GLM. And that's how

173
00:06:37,879 --> 00:06:41,360
multi-modal

174
00:06:39,800 --> 00:06:43,680
models and and setups should work,

175
00:06:41,360 --> 00:06:45,439
matching the task to the model, not just

176
00:06:43,680 --> 00:06:46,840
defaulting to one provider for

177
00:06:45,439 --> 00:06:48,680
everything because some of these are

178
00:06:46,839 --> 00:06:50,159
quite overpowered if that makes sense

179
00:06:48,680 --> 00:06:52,439
and expensive, too. And it's better to

180
00:06:50,160 --> 00:06:54,360
route your tasks to cheaper APIs that

181
00:06:52,439 --> 00:06:56,399
are way easier to use. And if you're

182
00:06:54,360 --> 00:06:59,000
trying to figure out how to actually set

183
00:06:56,399 --> 00:07:00,839
OpenClaw up, run it reliably, and use it

184
00:06:59,000 --> 00:07:02,319
for real business workflows, come join

185
00:07:00,839 --> 00:07:04,159
us in the Air Profit Bootcamp. We've got

186
00:07:02,319 --> 00:07:06,839
a 30-day roadmap built specifically

187
00:07:04,160 --> 00:07:09,080
around OpenClaw setup, agent design, how

188
00:07:06,839 --> 00:07:11,319
to use the {forward slash} tasks board,

189
00:07:09,079 --> 00:07:13,359
how to structure cron jobs and schedule

190
00:07:11,319 --> 00:07:15,439
tasks, how to set up a failover so your

191
00:07:13,360 --> 00:07:16,879
automations don't loop out at 3:00 a.m.,

192
00:07:15,439 --> 00:07:18,879
daily tutorials walking you through

193
00:07:16,879 --> 00:07:20,519
every update as it ships, and four

194
00:07:18,879 --> 00:07:22,800
coaching calls every week with people

195
00:07:20,519 --> 00:07:24,879
who are deep in OpenClaw right now.

196
00:07:22,800 --> 00:07:26,800
There's 2,700 business owners in there,

197
00:07:24,879 --> 00:07:28,839
a lot of them already running OpenClaw

198
00:07:26,800 --> 00:07:30,840
for their business or for client work,

199
00:07:28,839 --> 00:07:32,679
for lead gen, for content, for research.

200
00:07:30,839 --> 00:07:34,679
Link in the comments description or just

201
00:07:32,680 --> 00:07:37,360
go to the airprofitbootcamp.com to get

202
00:07:34,680 --> 00:07:39,000
access. So, there's also over 40 sick

203
00:07:37,360 --> 00:07:40,920
fixes like I was talking about, and

204
00:07:39,000 --> 00:07:43,639
here's what to actually pay attention

205
00:07:40,920 --> 00:07:46,439
to. Beyond the new features, 4.1 is a

206
00:07:43,639 --> 00:07:48,079
serious stability release, right? And

207
00:07:46,439 --> 00:07:51,120
let me flag the ones that matter most

208
00:07:48,079 --> 00:07:53,
}