Repository navigation
docs: add litellm.harness agent harness docs - #1945
Merged
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
3 of 5 tasks
# Conflicts: # src/css/custom.css
This branch was successfully deployed
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds docs for
litellm.harness, which runs Claude Code, Codex, OpenCode and Deep Agents from Python with one API. The SDK PR is still in progress in BerriAI/litellm (branchlitellm_harness_sdk), so the section is marked Beta and the API may still shift.The gateway is the main path. The overview example uses
Gateway(api_base=..., api_key=...)(orLITELLM_PROXY_API_BASE/LITELLM_PROXY_API_KEY), and the second page, "Using with AI Gateway", walks through aconfig.yamlwith onecodermodel group, creating a virtual key, running all four harnesses against it, and finding spend per harness through theharness,<name>request tags in the UI,/tag/daily/activityand/spend/logs. SDK-only mode is covered after that as the alternative.Capabilities match what v1 actually ships: runtimes have to already be in the sandbox, default permissions are
full, Claude Code and OpenCode support read-only/edit/full, Codex supports read-only/full with no tool filtering, and only Deep Agents has custom tools and history.What changed:
docs/harness/*.mdadds 17 pagessidebars.jsadds a new "Agent Harnesses" category after "LiteLLM AI Gateway (Proxy)" with non-collapsible Getting started / Harnesses / Guides / Reference groups, plus a link from the Integrations "Agent SDKs" list. No other entries change.src/css/custom.cssadds an IBM Plex / JetBrains Mono font import and a commentedlitellm.harness docsblock at the end. It only applies to.harness-sidebarand tohtml[class*='docs-doc-id-harness/'], so nothing else on the site is affected. Dark mode works.I built it in a local preview with the site's theme and the build passes. I checked the overview, the gateway page and the Codex page in light mode, and the supported-harnesses page in dark mode. The build output includes the uppercase group labels, the thin left rule with an accent bar on the active item, and the tables with light borders.