TokenHub is available as a Docker image, Linux package, and offline installer Download now
TokenHubAI Compute Orchestration Hub
中文 Free Download
Changelog

Shipping steadily. Better every release.

New features, improvements and fixes in every TokenHub release. TokenHub runs on your own servers, so you decide when to upgrade. Back up your data directory first.

SampleThe versions and dates below are sample content that shows the changelog format. Official release notes will reflect actual versions.

Releases

8 releases

  1. v2.4.0

    Latest

    Employee quotas · Model cost estimates

    New
    • Employee quotas: one key per employee, with expiry and allowed hours, spend / token / request budgets, and a model allowlist. See Employee Quotas.
    • Quota templates by department and role. With directory sync, keys are issued on hire and revoked when someone leaves.
    • Usage by provider: input and output tokens plus model fees, broken down by provider, model and employee. Local models can carry an internal cost rate.
    • Model cost estimator: estimate monthly provider fees from headcount, usage and model mix, and get a suggested budget per person.
    Improved
    • New overage actions: fall back to a local model, or request a top-up.

  2. v2.3.0

    One-click local models · Semantic cache

    New
    • One-click local models: automatically discovers Ollama / vLLM servers on your network, registers their models in bulk and runs connectivity checks.
    • Semantic cache: similar questions are answered straight from cache. Enable it per project with a similarity threshold and TTL.
    • Model health panel showing per-channel availability, average latency and recent errors.
    Improved
    • Smart routing adds three presets: cost-first, latency-first and quality-first.
    • Lower time-to-first-token for streaming, so long conversations feel smoother.
    Fixed
    • Tables spanning multiple pages in some PDFs were parsed out of order.
    • Live Dashboard legends overlapped at 4K resolution.

  3. v2.2.1

    Stability fixes

    Fixed
    • Users were occasionally sent back to the sign-in page after SSO login.
    • Time zone drift in usage reports spanning month boundaries.
    • The Rerank endpoint returned an error when top_n exceeded the number of documents.

  4. v2.2.0

    Hybrid RAG retrieval and pinpoint citations

    New
    • Hybrid RAG retrieval: fuses vector search with keyword search (BM25) and supports a rerank model for a second pass.
    • Source citations link to the exact page and paragraph, with a click-to-preview of the original.
    • Knowledge base hit rate and no-answer rate now appear on the Live Dashboard.
    Improved
    • Chunking can split by heading level to keep section context.
    • Bulk uploads resume after interruptions, making large-file ingestion more reliable.
    Fixed
    • Word headers and footers were ingested more than once.

  5. v2.1.0

    SSO and audit log export

    New
    • Single sign-on: OIDC, SAML 2.0 and LDAP, with automatic sync of departments and members.
    • Audit export: filter audit logs by time, member and model, then export to CSV or JSON.
    • Custom roles that combine permissions at the menu and action level.
    Improved
    • API keys support expiration dates and IP allowlists.
    • Bulk member import and bulk team assignment.
    Fixed
    • Keys linked to a deleted project could still make calls for a short time.

  6. v2.0.0

    Major release

    Live Dashboard and cost center

    New
    • Live Dashboard: real-time calls, token usage, cost, latency and error rates, optimized for 1920×1080 screens.
    • Cost center: department and project cost rankings with budget tracking and threshold alerts.
    • Full-screen mode and automatic rotation across dashboard pages.
    Improved
    • Redesigned admin console navigation with shorter paths to common tasks.
    • Call log search combines filters for model, status code and latency range.

  7. v1.5.0

    Multi-level permissions and quotas

    New
    • Three-tier RBAC across organizations, teams and projects.
    • Token quotas and rate limits per API key and project.
    • Embeddings and Rerank endpoints.
    Improved
    • Failover supports custom retry counts and fallback model order.
    Fixed
    • Streaming responses were occasionally cut off on unstable networks.

  8. v1.0.0

    First stable release

    Unified API gateway launches

    New
    • OpenAI-compatible Chat Completions endpoint with streaming.
    • Support for major cloud model providers and local models via Ollama.
    • Basic RAG Knowledge Base: upload, parse, chunk and vector search.
    • Call logs, usage stats and API key management.
    • Docker image and Linux package for self-hosted deployment.

Get the latest version

Download the newest TokenHub for free, run it on your own servers, and try one-click local models and hybrid RAG retrieval.

Free Download Read the Quickstart