Login Book a demo

Changelog

What’s new in Ollive

Release notes by email, when we ship. Privacy policy

Release 10

Ask Ollive in natural language

Ask about your agents from Codex, Claude, Gemini and other AI assistants, or build Ollive into your own systems.

NewAPI & MCP

Ask your AI assistant about your agents

Add https://mcp.ollive.ai/mcp to Codex, ChatGPT, Claude, Claude Code, Cursor or Gemini CLI and approve access. No API key needed. Then ask which rules your agents break, why a run was flagged, or how results are trending.

The assistant reads findings, results and runs with your permissions, and can send runs for checking.

Read the docs

NewAPI & MCP

Build on the Ollive API

Manage integrations, start monitoring agents, pull findings, results and metrics, and send runs for checking.

Admins create API keys in the new Admin section, each scoped to one workspace.

Read the docs

Other improvements

  • SecurityAccount & access

    API keys are shown once, expire after a set time duration, and any admin can revoke them.

  • SecurityAPI & MCP

    The approval screen shows where an app will redirect you and warns about apps Ollive can't verify.

Release 9

A new Home

Home opens on what needs attention, and the app uses an intuitive taxonomy throughout.

ImprovedHome

Start with what needs attention

See flagged runs over time, the agents and rules flagged most, and the agents that need review. Charts cover up to 180 days, and time ranges default to All time.

Release 8

Spend less time sorting findings

Filter findings, read each in one line, and export the list to Excel.

ImprovedFindings

Filters, summaries and Excel export

Filter by agent, rule set, severity and time. Each finding gets a one-line summary and a count of the runs and agents affected, and opens on the first failed step.

Export the filtered list to Excel, one row per affected run, each linked back to Ollive.

Other improvements

  • ImprovedFindings

    Evidence reads as questions and answers, with long sections collapsed.

  • ImprovedIntegrations

    The connect screen spells out how your data is handled: read-only access, encrypted credentials and no model training.

Release 7

Broken rules, flagged in Slack

Get a Slack message when a monitored agent breaks a rule.

NewMonitoring

Slack alerts

Connect Slack from Integrations, pick a channel and send a test message. Then turn on alerts from an agent's page.

Each alert names the rule, its severity and its source, and links straight to the agent in Ollive.

Release 6

Seven new rule sets, plus your own

Seven rule sets join the library, from HIPAA to the OWASP LLM Top 10, and we build custom ones on request.

Platform PacksRule sets

HIPAA, TCPA, FDA, MITRE ATLAS and OWASP

New in the library: HIPAA, TCPA rules for AI voice calls, three FDA rule sets (drug promotion, biologics safety reporting, and product complaints and quality), MITRE ATLAS and the OWASP Top 10 for LLM Applications. Turn on the ones that apply and Ollive checks your agents against them.

NewRule sets

Your policies and contracts, as rules

Ask us to add a regulation to the library, or to build a rule set from your internal policies. Upload a customer contract and we turn its commitments into rules your agents are checked against.

Other improvements

  • ImprovedIntegrations

    Retell calls are checked as soon as they end.

  • ImprovedMonitoring

    Long runs are checked end to end, with nothing cut off.

Release 5

Findings, traced to the law

See every finding on one page, each linked to its rule and the legal text behind it.

NewFindings

One page for all your findings

Findings lists every rule your agents have broken, across all agents. Repeat failures roll up into one finding with a run count, and the runs inside are grouped by how the rule was broken.

NewFindings

Evidence behind every flag

Each finding shows the rule, the legal clause and citation behind it, and a link to the source, as they stood when the run was checked. Later rule updates don't change past results.

Evidence shows what each answer relied on: the run, or the product and company details you gave Ollive. The transcript marks the turn or tool call where the agent went wrong.

Other improvements

  • ImprovedAgents

    The Agents page groups agents by platform and shows run counts for the time window you pick.

Release 4

Live runs, checked against the EU AI Act

Ollive checks each live run against the rule sets you pick, starting with the EU AI Act.

NewMonitoring

Rule-by-rule monitoring

Turn on monitoring for an agent and Ollive checks each of its runs against the EU AI Act, one rule at a time. Flagged runs link straight to the full conversation.

Platform PacksRule sets

Set up in three steps

Connect a platform, choose your rule sets and pick the agents to monitor. The rule set library shows the rules inside each regulation and framework.

Other improvements

  • FixedAgents

    Agents built with LangGraph and other graph frameworks show under their own names, including those found in past runs.

Release 3

Connect the platforms your agents already use

Ollive reads runs straight from Braintrust, Langfuse, LangSmith, Cekura and Retell.

NewIntegrations

Five new integrations

Connect Langfuse, LangSmith or Braintrust for the agents you build, or Retell and Cekura for voice agents. Ollive reads each run step by step, including what each tool was sent and what it returned.

Release 2

Failed tests, grouped by what went wrong

See each agent's pass rate and the problems behind its failed tests.

ImprovedAgents

From pass rate to conversation

Smart grouping for failed tests based on the issues behind them, next to the agent's pass rate. Open an issue to see its tests, then open a test to read the conversation.

Other improvements

  • NewAgents

    Custom agents can be added manually when their platform isn't connected.

  • ImprovedGeneral

    The app has a new look and simpler navigation.

Release 1

Test your AI agents before they go live

Import voice agents from Retell and catch their failures before launch.

NewAgents

Pre-launch tests for voice agents

Import a voice agent from Retell, describe how it is used, and send it for testing. After a change, re-run the tests without describing it again.

Failed tests show first on the agent's page.

Other improvements

  • NewHome

    Home lists the agents that need attention before launch.

Day one

Ollive came into being

Take on bigger AI deployments
without taking on unlimited AI risk.

Book a demo