> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getbifrost.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# v1.5.16

> v1.5.16 changelog - 2026-06-21

<Tabs>
  <Tab title="NPX">
    ```bash theme={null}
    npx -y @maximhq/bifrost --transport-version v1.5.16
    ```
  </Tab>

  <Tab title="Docker">
    ```bash theme={null}
    docker pull maximhq/bifrost:v1.5.16
    docker run -p 8080:8080 maximhq/bifrost:v1.5.16
    ```
  </Tab>
</Tabs>

<Update label="Bifrost(HTTP)" description="1.5.16">
  ## ✨ Features

  * **Failed-Request Billing** — Bill for tokens a provider already processed when a stream is cancelled or times out, so partial usage is still accounted (closes #3357)
  * **Airgapped Local Sync** — Model catalog pricing and model parameters can now sync from local files for airgapped deployments (closes #4305)
  * **MCP Extra Header Forwarding** — Allowlisted per-request extra headers are now forwarded to MCP tool calls, including `ping`/`list_tools`
  * **Business Unit & User Tracing** — Traces now capture business unit and user names/IDs for richer attribution
  * **Dashboard Export** — Added a dashboard export endpoint to download dashboard data
  * **Virtual Key Rankings** — New Virtual Key Rankings tab in the dashboard
  * **Group Traces by Sessions** — Traces can now be grouped by session via config.json and Helm
  * **Guardrails Evaluation Mode** — Exposed evaluation mode in guardrail schemas

  ## 🐞 Fixed

  * **Anthropic Streaming** — Fixed duplicate `message_start` event in the Anthropic stream (closes #4556)
  * **Bedrock Streaming** — Encode Bedrock stream errors as EventStream exceptions, fixing ChecksumMismatch / corrupted EventStream on PostLLMHook errors (closes #4545) (thanks [@jstar0](https://github.com/jstar0)!)
  * **Bedrock MCP Tools** — Strip provider-unsupported server tools (e.g. `mcp`) on the Bedrock/Anthropic Responses path instead of failing the whole request, restoring pre-v1.5.0 behavior (closes #3795)
  * **Bedrock Streaming Buffering** — Send `Accept-Encoding: identity` on Bedrock streams so responses arrive incrementally instead of buffering until generation completes (fixes TTFB collapse) (closes #4542)
  * **OpenRouter Server Tools** — Preserve OpenRouter server tools (`openrouter:web_search`) on the OpenAI Responses path instead of stripping them in `filterUnsupportedTools` (thanks [@abdenasseraroukhsiss](https://github.com/abdenasseraroukhsiss)!)
  * **Vertex Image URLs** — Remote image URLs are now inlined to base64 for Anthropic-on-Vertex, which rejects url-source images (previously only documents were inlined)
  * **Bedrock MiniMax** — Fixed Bedrock signature handling for MiniMax models
  * **Bedrock Nova** — Fixed Nova model handling on Bedrock
  * **Bedrock Batch** — Corrected model id in Bedrock batch requests
  * **Feature Gating** — Resolve model names for feature gating; return 403 correctly and fix 403 errors on list-models requests
  * **Governance on OAuth** — Run governance on Claude Code OAuth requests when a virtual key is present; removed the skip-key-selection check from the pre-LLM hook
  * **Conflict Handling** — Return 409 for Conflict errors
  * **Container Delete** — Added explicit content-type header for container delete
  * **File Serving** — Handle URL-encoded file names in URL params for the single-file serving endpoint
  * **Plugins Config** — Fixed redaction setting in plugins config (#4486)
  * **Migrations** — Fixed canonical\_model\_view migration

  ## 🔧 Maintenance

  * **Network Defaults** — Updated default network config timings
  * **Dependencies** — Dependabot dependency updates

  ## 🐙 Closed GitHub Issues

  * [#2887](https://github.com/maximhq/bifrost/issues/2887) — Listing models for a virtual key not associated with all providers adds errors to logs
  * [#3357](https://github.com/maximhq/bifrost/issues/3357) — Bifrost Billing Discrepancy Analysis Report
  * [#3795](https://github.com/maximhq/bifrost/issues/3795) — MCP tools fail with Bedrock provider in v1.5.0
  * [#4305](https://github.com/maximhq/bifrost/issues/4305) — Model parameters can't be configured locally
  * [#4530](https://github.com/maximhq/bifrost/issues/4530) — /v1/responses strips OpenRouter server tools (openrouter:web\_search) via filterUnsupportedTools
  * [#4542](https://github.com/maximhq/bifrost/issues/4542) — Bedrock streaming responses buffered server-side — TTFB matches total generation time
  * [#4545](https://github.com/maximhq/bifrost/issues/4545) — Bedrock streaming ChecksumMismatch / corrupted AWS EventStream
  * [#4556](https://github.com/maximhq/bifrost/issues/4556) — Duplicate `message_start` SSE frame on Anthropic passthrough
</Update>

<Update label="Core" description="1.5.22">
  * fix: inline remote image URLs (not just documents) to base64 for Anthropic-on-Vertex, which rejects url-source images
  * fix: preserve OpenRouter server tools (openrouter:web\_search) in filterUnsupportedTools on the OpenAI Responses path (thanks [@abdenasseraroukhsiss](https://github.com/abdenasseraroukhsiss)!)
  * fix: send Accept-Encoding: identity on Bedrock streaming requests so responses stream incrementally instead of buffering until completion (closes #4542)
  * feat: bill partial usage for cancelled or timed-out streaming requests via BilledUsage on errors (closes #3357)
  * feat: forward allowlisted per-request extra headers to MCP tool calls, including ping/list\_tools
  * fix: strip provider-unsupported server tools (e.g. mcp) on the bedrock/anthropic responses path instead of failing the request (closes #3795)
  * feat: add business unit & user names and ids to tracing
  * fix: double message start event in anthropic stream (closes #4556)
  * fix: bedrock signature fix for minimax
  * fix: nova bedrock fixes
  * fix: model id in bedrock batch request
  * fix: resolve model names for feature gating
  * fix: 403 errors for list models request
  * fix: explicit content type header for container delete
  * chore: updated the default network config timings
</Update>

<Update label="Framework" description="1.3.22">
  * feat: sync model catalog pricing and model parameters from local files for airgapped deployments (closes #4305)
  * feat: compute cost for failed and cancelled requests from billed usage
  * feat: dashboard export endpoint
  * feat: add Virtual Key Rankings tab in dashboard
  * feat: add business unit & user names and ids to tracing
  * fix: return 409 for Conflict errors
  * fix: canonical\_model\_view migration
  * chore: upgraded core to v1.5.22
</Update>

<Update label="compat" description="0.1.21">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="governance" description="1.5.22">
  * feat: account billing for failed and cancelled requests
  * fix: remove skip key selection check from governance pre llm hook
  * fix: run governance on cc oauth requests when virtual key is present
  * fix: 403 errors for list models request
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="jsonparser" description="1.5.22">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="logging" description="1.5.22">
  * feat: log cost for failed and cancelled requests
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="maxim" description="1.6.22">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="mocker" description="1.5.22">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="modelcatalogresolver" description="1.0.3">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="otel" description="1.2.23">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="prompts" description="1.0.22">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="semanticcache" description="1.5.22">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>

<Update label="telemetry" description="1.5.22">
  * chore: upgraded core to v1.5.22 and framework to v1.3.22
</Update>