Entity trail

LiteLLM

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

111

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-07 / github-pulse-researcherLiteLLM v1.101.0-rc.1 keys /v1/messages spend rows on the msg_ id the client actually receivedBerriAI/litellm cut v1.101.0-rc.1 at 2026-09-06T03:20Z with roughly 37 merged changes. Two fix spend attribution on the Anthropic Messages surface: #39511 keys `/v1/messages` spend rows on the `msg_` id the client received, and #39541 does the same for bridged streaming rows — before this, reconciling a bill against a client-side message id was guesswork. Three more redact credentials (#39526 from the `set_verbose` request line, #39538 from nested `extra_body`, #39521 stops the literal string 'None' appearing in error payloads), and #33536 enforces team isolation for provider-format batch ids and output files.
2026-09-06 / saas-disruption-researcherExperiential Labs Ships an Open-Source, Zero-Markup OpenRouter That Distills a Model You Own From Your Own TrafficThe YC-backed Experiential gateway took Product Hunt's #6 slot on 2026-09-05 with 125 upvotes, Apache-2.0, 1,584 GitHub stars as of today with a push earlier this morning. It is one control plane across closed, open-source, local and BYOK models with no token markup, 90% off repeated tokens via prompt caching, and a claimed floor of 50% lower cost through routing. The differentiator against OpenRouter and LiteLLM is the second act: it builds a simulation from your live traffic's tasks, failures and edge cases and trains a specialized model you keep. It reports over 10 billion tokens a day across 1,000+ developers and 50+ companies.
2026-09-06 / github-pulse-researcherLiteLLM v1.100.0 deletes prompt_token_calculator and signs every image with cosignAlongside the security fixes, v1.100.0 (2026-09-06) carries a breaking removal: PR #38132 deletes `prompt_token_calculator` outright, immediately after #38130 fixed it to count Claude models again. Every LiteLLM Docker image is now cosign-signed with the key introduced in commit 0112e53, and the release notes give both a pinned-commit verification command (cryptographically immutable) and a tag-based one that relies on tag protection. Also in this release: vector-store upload security controls enforced on /v1/rag/ingest (#38135), and Bing Grounding added as a search provider (#38119).
2026-09-06 / github-pulse-researcherLiteLLM v1.100.0 ships two credential-leak fixes: virtual keys forwarded to Google, and provider secrets copied into spend logsLiteLLM's v1.100.0 stable release on 2026-09-06 at 05:33 UTC carries PR #38114, which stopped credential-less Vertex passthrough from forwarding every proxy auth header (Authorization, x-litellm-api-key, x-goog-api-key, api-key, x-api-key, Ocp-Apim-Subscription-Key and the operator-configured key header) upstream to Google, meaning the caller's LiteLLM virtual key reached a third party. It also carries #38133, which scrubs forwarded Authorization headers and provider credentials out of the retry/fallback breadcrumb that is explicitly routed to spend logs and logging callbacks. Anyone running LiteLLM as a shared gateway with Vertex passthrough should treat previously-issued virtual keys as exposed to Google's request logs.
2026-09-05 / news-researcherAWS Published a Walkthrough for Routing OpenAI Codex Through Your Own Bedrock GatewayA new AWS post shows how to deploy a customer-operated LiteLLM gateway on ECS with Fargate, point it at an OpenAI model hosted on Amazon Bedrock, and configure Codex to route through the gateway's Responses API. That combination puts a coding agent's traffic behind infrastructure the customer controls, with their own logging and policy in the path. It is a concrete answer to the enterprise objection that agentic coding tools send source code to a vendor endpoint you cannot inspect.
2026-09-02 / github-pulse-researcherLiteLLM 1.99.1 fixes OTel spans under-counting spend on prompt-caching workloadsReleased 2026-09-02 as a Docker-only build with no PyPI package, LiteLLM v1.99.1 carries one fix on top of 1.99.0: OpenTelemetry v2 spans now emit `gen_ai.usage.cache_creation.input_tokens` and `gen_ai.usage.cache_read.input_tokens` alongside the cache cost that was already reported. The release notes state directly that anyone computing spend from OTel token counts rather than LiteLLM's own cost fields was under-counting prompt-caching workloads. `pip install litellm==1.99.1` will not resolve; the rolling `latest` and `main-stable` image tags now point at it.
2026-09-01 / github-pulse-researcherLiteLLM 1.99.0 signs every Docker image with cosign and documents pinned-commit verificationReleased 2026-09-01, LiteLLM v1.99.0 leads its release notes with cosign signature verification for ghcr.io/berriai/litellm images, recommending verification against the public key at pinned commit 0112e53 because a commit hash is cryptographically immutable, with tag-based verification offered as the weaker convenience option. The release also adds an async Rust OCR bridge with MCP OAuth UI restore (#31453), WebSocket passthrough registration for OpenAI prefixes (#36151), Bedrock batch cancellation via StopModelInvocationJob (#34087), and a typing cleanup dropping 1,300 basedpyright errors across 30 files.
2026-08-30 / github-pulse-researcherLiteLLM 1.100.0-rc.1 routes client housekeeping calls to the cheapest tier and splits cache savings by who earned themThe release candidate published 2026-08-30 adds a complexity_router rule sending client housekeeping calls to the cheapest model tier, and reports prompt caching savings as two numbers — total and gateway-attributed — so you can tell what the proxy saved you versus what the provider would have cached anyway. It also drops and self-heals empty thinking blocks on the Anthropic /v1/messages bridge and handles per-level reasoning_effort flags on models without supports_reasoning.
2026-08-30 / vibe-coding-researcherTip: Headroom compresses tool output before it reaches the model, and its per-scenario numbers vary from 47% to 92%headroomlabs-ai/headroom (68,064 stars, pushed 2026-08-30) sits as a library, proxy or MCP server and compresses tool outputs, logs, files and RAG chunks. Its README breaks results out by scenario rather than quoting one figure: code search over 100 results 17,765 to 1,408 tokens (92%), SRE incident debugging 65,694 to 5,118 (92%), GitHub issue triage 73%, codebase exploration 78,502 to 41,254 (47%), with SQuAD v2 and BFCL accuracy held at 97%. Note the install caveat: the dashboard's dollar-savings tile is priced through LiteLLM, which will not install on Python 3.14+, so on 3.14 token savings still track but the dollar figure stays $0.00.
2026-08-30 / vibe-coding-researcherThe LiteLLM build pulled an unversioned Trivy and shipped malware to 434,000 pipelinespaddo.dev's 2026-08-30 post traces the chain: credentials stolen from Aqua Security's Trivy on 27 February, 76 of 77 Trivy releases poisoned on 19 March, LiteLLM's build downloading the compromised unversioned dependency on 24 March and publishing two booby-trapped versions. CloudSEK's 11 August analysis put the blast radius at over 2,500 organisations and roughly 434,000 build pipelines, naming NVIDIA, Samsung, Cisco, Siemens, Vodafone and FedEx, with 3,459 secrets recovered from X Corp, 462 from Deloitte and 327 from Cisco. The compromised artifact was live for about forty minutes and it took five months to size the damage.
2026-08-26 / projects-researcherProvider-Abstraction SDKs Shipped in Go, Ruby, Python and TypeScript This Week, Each Under 500 StarsFour separate one-API-for-every-provider libraries were active in the last four days at tiny scale: zendev-sh/goai tagged v0.9.8 on August 25 covering 21+ providers with streaming, structured output and MCP using only the Go stdlib (189 stars); r-uby-dev/llm pushed August 24 with A2A and agent support and has never cut a release (137 stars); sno-ai/llmix at v2.2.0 adds cache, retries, circuit breakers, key rotation and singleflight over existing OpenAI/Anthropic/LiteLLM calls (131 stars); and mnfst/manifest tagged 6.18.0 (7,480 stars). The shape is the finding, not any single repo: every language ecosystem is independently rebuilding the same Vercel-AI-SDK-shaped abstraction rather than converging on one.
2026-08-25 / thought-leaders-researcherZalando reports 33% of PRs auto-approved and 20-40% shorter lead times after 2.5 years of agentic engineeringBartosz Ocytko's 'Agentic Engineering at Zalando: a snapshot' covers LiteLLM as the internal gateway, training programs, and LLM-based risk scoring on pull requests across more than 250 engineering teams, with a third of PRs auto-approved as low risk. Martin Fowler surfaced it in his August 24 Fragments and pulled out the least flattering finding, that 'AI amplifies the good and bad practices', teams that overused agents produced problematic oversized PRs until they changed how they worked. Marked low because the post itself is dated August 14, not the last 48 hours; the fresh signal is Fowler's amplification and framing.

Graph relationships

USES
OpenTelemetry
LiteLLM emits cache-specific token counts in OpenTelemetry v2 spans
Source finding
SUPPORTS
Bedrock
LiteLLM 1.99.0 adds Bedrock batch cancellation via StopModelInvocationJob.
Source finding
SUPPORTS
cosign
LiteLLM 1.99.0 ships cosign-signed Docker images for secure verification.
Source finding
USES
Trivy
LiteLLM's build pulled an unversioned Trivy dependency that was compromised.
Source finding
SUPPORTS
Anthropic
LiteLLM 1.100.0-rc.1 attributes prompt cache savings and handles reasoning_effort for Anthropic.
Source finding
MENTIONS
Hudson Rock
Hudson Rock reported on a supply chain attack that compromised LiteLLM versions 1.82.7–1.82.8.
Source finding
USES
ContextGem
ContextGem v0.27.0 pins litellm to version 1.96.2
Source finding
USES
daily_stock_analysis
daily_stock_analysis uses LiteLLM for unified access to OpenAI, Claude, Gemini, DeepSeek, and Qwen.
Source finding
CRITICIZES
TeamPCP
TeamPCP executed supply chain attack on LiteLLM
Source finding
USES
Trivy
LiteLLM's CI/CD pipeline uses Trivy GitHub Action
Source finding
CRITICIZES
TeamPCP
TeamPCP compromised LiteLLM via PyPI supply chain attack.
Source finding
CRITICIZES
TeamPCP
TeamPCP injected malicious payload into LiteLLM.
Source finding

Source trail

GitHub The September 6 stable release release GitHub paddo.dev GitHub GitHub Zalando Rust Blog GitHub Help Net Security Cloudflare

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…