Repository navigation
Update the sandbox limit to 25, document self-service Hermes updates and rewrite the CrewAI guide for the built-in provider - #40
Merged
Conversation
Hosted Hermes-Agent instances are no longer upgraded in place by the platform: the customer connects over SSH and runs the update command that the dashboard's Update action shows. Add an Updating section with that command, worded like the dashboard, and note that the version shown in the dashboard does not change after a self-update.
The Update action still upgrades OpenClaw in place, but for Hermes-Agent it now shows the SSH and update commands to run instead. Say so in the Hosted Agents introduction and link the Hermes-Agent steps.
The backend refuses to update Hermes-Agent instances in place and answers 409 with the command to run over SSH. Say so under the HTTP API example.
crewai 1.15.27 (released 2026-10-09) ships DeepInfra as a native OpenAI-compatible provider (crewAIInc/crewAI#7459), so LLM(model="deepinfra/<org>/<model>") reads DEEPINFRA_API_KEY and needs neither a base URL nor LiteLLM. - Configuration collapses to the one-line LLM(model="deepinfra/...") in all three examples; DEEPINFRA_BASE_URL is documented for proxies. - Drop the @llm workaround note: a plain `llm: deepinfra/...` string now works in agents.yaml, so show that snippet instead. - Pin the install line to crewai>=1.15.27 and bump the "Tested with" note. - Tips: replace the LiteLLM tip with an "Older CrewAI versions" fallback (custom_openai=True + base_url), add the org/model id check that now raises a ValueError at construction, and keep the context-window tip since the native provider still assumes 8,192 tokens for DeepInfra ids. The embeddings section keeps the `openai` embedder pointed at DeepInfra: the native DeepInfra embeddings provider (crewAIInc/crewAI#7978) is not merged yet. Every Python block in the page was run verbatim against api.deepinfra.com with crewai 1.15.27, including the YAML crew path, structured outputs, streaming and the knowledge example.
smanilov
approved these changes
Oct 12, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR carries three independent doc updates.
Sandbox limit: 5 → 25
The per-account cap on active sandboxes goes from 5 to 25. This updates the three places that stated the old number:
sandboxes/overview.mdx, Limits and quotas: "up to 25 non-stopped sandboxes per account"sandboxes/overview.mdx, Errors table: "Over the 25-active-sandboxes cap"integrations/crewai.mdx: "Note the account limit of 25 active sandboxes"Only the number changes. Going over the cap still returns a
429(RateLimitError/TooManySandboxesError), and stopped sandboxes still don't count toward it.A search of the whole repo, including
api-reference/openapi.json, found no other place that states the limit.Hosted Hermes-Agent: self-service updates
The platform no longer upgrades Hermes-Agent instances in place. The customer connects over SSH and runs the update command that the dashboard's Update action shows. Three places described the old flow, and each fix has its own commit:
agents/hermes-agent.mdx(28146a5). It gives the SSH steps and the update command, worded like the dashboard dialog, and links the upstream releases. It suggests a backup first, sincehermes updatepulls upstreammain. It also notes that the version shown in the dashboard doesn't change after a self-update, and thathermes --versionshows what is running.agents/introduction.mdx, Manage an instance (3634067). It said every framework upgrades in place. It now says that applies to OpenClaw, and that for Hermes-Agent Update shows the commands to run.agents/introduction.mdx(4f4d5f1).POST /v1/agents/{instance_id}/updatenow returns409for Hermes-Agent instances, with the command to run. A line under the example says so.The command is written out in full (
git -C /opt/hermes/src remote set-branches origin main && hermes update) rather than as a barehermes update. Thegitpart is needed on older instances that haven't been restarted since the change, and it is harmless on the rest.These stay unchanged because they are still correct:
agents/openclaw.mdx: OpenClaw still upgrades in place.api.deepinfra.com/openapi.json, so it picks up the409on its own.CrewAI: built-in DeepInfra provider
crewai 1.15.27 (2026-10-09) ships DeepInfra as a native OpenAI-compatible provider (crewAIInc/crewAI#7459):
LLM(model="deepinfra/<org>/<model>")readsDEEPINFRA_API_KEYand needs neither a base URL nor LiteLLM.integrations/crewai.mdxis rewritten around that (9958ffe):LLM(model="deepinfra/deepseek-ai/DeepSeek-V4-Flash-0731")in all three examples, replacing thecustom_openai=True/base_url/api_keyblock.DEEPINFRA_BASE_URLis documented for proxies.@llmworkaround note is removed. A plainllm: deepinfra/...string now works inagents.yaml, so the page shows that snippet instead.crewai>=1.15.27; the "Tested with" note moves from 1.15.21 to 1.15.27.custom_openai=True+base_url, which still works on current versions); a new tip covers theorg/modelid check, which now raises aValueErrorat construction; the context-window tip stays, because the native provider still assumes 8,192 tokens for DeepInfra ids.Unchanged on purpose:
openaiembedder withapi_basepointed at DeepInfra. The native DeepInfra embeddings provider (crewAIInc/crewAI#7978) is not merged yet; this section should switch to it once it ships.Verification: every Python block in the page was extracted and run verbatim against
api.deepinfra.comwith crewai 1.15.27, plus a@CrewBasecrew configured only throughagents.yaml, structured outputs (response_formatandoutput_pydantic), streaming, and theDEEPINFRA_BASE_URLoverride.Not in this PR
The fleet-wide cap is also going up, but the docs never give that number, so the Fleet capacity bullet is unchanged.