LineMod release history

2026-09-20 — After a content hit you can reply instead of only blocking. Tool protection follows what the call is doing. Pictures can be checked, and the catalog adds image generation plus newer text models.

• Reply after a content hit: choose a short refusal or your own reply text, or keep blocking the request. Try the reply in Console before it goes live. See the security guide.
• Tool protection: mark tools as trusted or blocked. Protection looks at the operation (for example a delete action), not everyday words in a sentence.
• Security workspace: approvals live with tool security. Events stay on the same trace as the run, and you can see why a run was flagged.
• Personal information: fewer false flags on phone numbers and ID numbers. Card checks are tighter. Web markup or database commands in a model reply are no longer treated as an attack.
• Images and models: check readable text in pictures you send. Generate images through the same gateway. New models include GLM-5.3, DeepSeek Flash, and GPT-6 Astra.
• Sign-in and access: extra check that you are a person when sign-in looks risky. New accounts can join a request-access waitlist.

2026-08-05 — See what each agent spent, where tokens are wasted, and drill from a summary down to individual steps.

• Agent insights: compare spend, security events, and per-run cost by agent type in Console — filter the table and export CSV.
• Run health diagnostics: highlights whether cache is helping, tool definitions are eating context, steps are idling, and where money goes.
• Run detail notes: on a run waterfall, view baseline comparison and steps that may need attention.
• Clearer attribution: integrations can declare the agent type and link every step in a multi-turn run. See the SDK guide.
• Docs and site: quickstart, SDK, and observability pages now explain agent observability.

2026-07-20 — First production launch of LineMod on linemod.ai and api.linemod.ai.

• OpenAI-compatible gateway: org-scoped API keys and streaming responses.
• Smart routing: auto:* model templates, org routing rules, failover, and route decision transparency in traces.
• Security guardrails: inbound prompt scanning, sensitive-data handling, and agent tool governance with audit evidence.
• Tenant console: onboarding, API keys, routing policy, security policy, audit runs, usage analytics, and billing paths (platform credits or BYOK subscription).
• Console agent mode: streamed assistant turns with grounding sources linked to audit records.
• Documentation hub (English and Chinese): quickstart, API reference, routing and security guides, SDK notes, and this changelog.
• Marketing site: routing, security, observability, pricing, and legal pages with light/dark themes.