What we’ve shipped

Skills & Knowledge GA, Danish-language audit reports

Agents can now write their own audit memos in Danish, ready for the DPO. Custom skills can be sealed in the ledger for review and shared across teams.

• Skills can be promoted from Personal → Team → Vault with one approval
• Danish-language audit report templates included
• Two new built-in skills: weekly_digest, stale_deal_radar

Email conversations, on-prem GPU support

Agents can now triage and draft email replies inside Gmail and Outlook with full conversation memory. On-prem Llama 3.3 deployment is now a one-line install.

• Gmail + Outlook native drafting
• On-prem Llama 3.3 70B via Docker Compose
• Inbox digest with priority routing

Article 14 evidence export

One command exports a quarter’s worth of agent decisions in the official EU AI Act Article 14 format. Your DPO will read it for an hour and sign.

• Quarterly Article 14 export bundle
• Signed PDF + JSON-LD pairs for every agent action
• Daily root hashes published at status.wildfire.io/chain

Model picker, Mistral Large in production

Change provider in one setting. Your agents keep working. Mistral Large is now the default for Tier 2 (client confidential) work.

• One-click model swap per agent or per tier
• Mistral Large via Paris region
• Gemini 2.5 added to the picker

Memory promotions, fewer false moves

Promoting a memory from Personal to Team now requires steward approval. We caught and rolled back four cross-team leaks during our internal pilot.

• Steward approval queue for memory promotions
• Auto-detection of regrettable share patterns

SOC 2 Type II completed

External audit closed, report available under NDA. ISO 27001 certification scheduled for Q2.