ContextGate 0.8.0
From an existing business query to a reusable Agent tool.
简体中文 · Publishing guide · Support demo · Installation
What's new
- One publishing workflow. Choose a source, provide a native query or fixed HTTP read operation, explain its inputs/results, run real checks, review the full source draft and publish. Save progress and recover from editing conflicts without overwriting unrelated entries.
- Concepts lead to queries. Link mapped ontology concepts by name, start a query from an entity, and inspect its definitions in the same editor. Published coverage separates linked queries from currently executable queries. Agents discover authorized concept/query links through the existing semantic tools.
- One query catalog. Available queries, administrator drafts, queries needing attention and all definitions have separate views. A query name opens its workspace; returning restores filters, the snapshot page, position and focus. Choosing an Agent removes management views, with the same boundary enforced by the API.
- Clear client confirmation. Connect an Agent, copy the tool call and wait for evidence. Waiting is bounded to two minutes and cancels when leaving or hiding the page, closing the panel, or changing the query/access. Confirmation matches the source, template, execution version and selected Agent; previews do not count.
- Less interface overhead. Home retains up to five account-scoped recent queries and highlights one next step. Contextual help, focused settings, compact rows, clear empty states, parameter examples and responsive ontology inspectors reduce navigation and repeated explanation.
Publication, execution checks, connection checks, health checks and business acceptance remain separate facts. A successful call is not a business-correctness verdict. The pilot worksheet provides a way to measure actual setup effort, answer quality and sustained reuse; no customer benefit is inferred from fixture tests.
Compatibility and upgrade
From 0.6.x or 0.7.x, back up the PostgreSQL metadata database, matching master key and deployment configuration. Stop ContextGate, replace the complete archive or image with 0.8.0 and restart with the same database URL, key and source-file directories. Verify a known query and audit record afterward.
This release adds no metadata migration or forced credential rotation. Administrator accounts, personal Configuration MCP identities, query Agent/OAuth grants, semantic publications, ontology mappings and old routes remain compatible. Semantic exports remain format v2 with v1 import support. There are still 15 query MCP tools and 22 Configuration MCP tools; query tools reuse execute_query_template, not one new MCP tool per template.
Saved browser resume records migrate to a bounded recent list when a query is opened. Storage contains only source/template/Agent IDs, scoped to the current administrator; query content, parameters, results and credentials remain excluded. A disabled browser store affects shortcuts, not saved server drafts.
From 0.4.x or 0.5.x, also follow the administrator migration checklist: sign in as admin with the retained password and issue personal Configuration MCP tokens to replace revoked legacy tokens. From 0.3.x or earlier, follow the PostgreSQL metadata migration guidance.
Packages and evidence
Download contextgate-0.8.0-linux-arm64.tar.gz or contextgate-0.8.0-linux-amd64.tar.gz and verify SHA256SUMS. Keep the launcher, libexec and private C++ libraries together. Both packages include the embedded UI, bilingual help, query/ontology/cloud examples, Apache-2.0 LICENSE/NOTICE and dependency notices. Linux glibc 2.36+ and PostgreSQL metadata are required; Node.js is not required at runtime.
The release provides release-verification.json, both dist-*.json and both download-*.json reports, identifying the exact commit, CI jobs, archive hashes and independently downloaded package checks. The package workflow includes PostgreSQL/HTTP API reads, HTTP/stdio, templates and ontology context, administrator recovery, revocation and both OTLP transports. Local amd64 execution uses OrbStack emulation; CI runs on native arm64 and amd64 runners.
See validation for each stage's actual scope. The full 18-product / 20-version matrix remains historical evidence; this release's package checks do not claim a new full-matrix run. Snowflake, Databricks SQL, BigQuery and Redshift remain preview connectors without real vendor verification. No federation, ontology instance store, inference engine, automatic query compilation or new resource-isolation model is introduced.