Changelog
Releases and product updates for Radiant Platform.
Upcoming0.12.0
Follow conversations across runs
Section titled “Follow conversations across runs”The new Conversations view brings related agent runs together across turns, including conversations handled by more than one agent. Find it between All agent runs and All traces in Logs. Runs are grouped by their conversation identifier within the selected time range.
Claude Code interactions and Codex turns now appear as agent runs, with agent identity and recorded usage. Their session identifiers connect related turns in Conversations.
Search and save Logs views
Section titled “Search and save Logs views”Logs searches, filters and sorts across the full result set. Use field and value suggestions to narrow the results, and arrange the columns for your investigation.
Save the query, time range, grouping and layout as a private or shared view. Stable links let you return to it later. Group summaries and details include the same related activity within the selected time range.
Hand off coding tasks
Section titled “Hand off coding tasks”Use /handoff to turn an investigation into a coding task. Your coding assistant picks it up through MCP; its questions and reported outcome stay in the original conversation.
Connect Slack and Vapi
Section titled “Connect Slack and Vapi”- Slack: mention Radiant in a channel or send a direct message to investigate in a thread, using your linked account’s project and tool access.
- Vapi: open a call span’s Recording tab to listen alongside its synchronized transcript, with waveform seeking and downloads.
MCP client update
Section titled “MCP client update”Breaking: update find_agent_runs and find_groups requests to use structured conditions and time ranges. inspect_group requires explicit time bounds. Group values now distinguish numbers, text and booleans.
0.11.0
Manage team access
Section titled “Manage team access”Invite teammates as members or admins, change roles, and remove access from Organization settings → Team. Organization roles separate everyday work, administration and ownership. Invitations check seat capacity when sent and accepted.
Share investigations
Section titled “Share investigations”Meta Agent conversations can stay private, be shared for reading, or accept teammates’ follow-up questions. Each participant uses their own tool connections and permissions and approves their own actions. Personal tool preferences can restrict the project’s policy.
Onboarding and invitation links now return you to the appropriate project, and dashboard controls update when your access changes.
0.10.3
Security maintenance
Section titled “Security maintenance”Dependency updates address Unix file-descriptor safety and cryptographic timing vulnerabilities.
0.10.2
Private-network simulation executors
Section titled “Private-network simulation executors”Private-network simulation executors can now be reached over HTTP when explicitly enabled. Connection addresses are checked before credentials are sent.
0.10.1
More reliable usage reporting
Section titled “More reliable usage reporting”If usage reporting is refused for one organization, reporting continues for the others. The affected records are retained and retried rather than interrupting the queue.
0.10.0
Guided project setup and account settings
Section titled “Guided project setup and account settings”New onboarding guides you through creating an organization and project, choosing a plan, and connecting your first trace.
Account settings now include profile and email changes, password changes, sign-out across devices and account deletion.
Plans and credit usage
Section titled “Plans and credit usage”Owners can manage subscriptions and payment details in Organization settings → Plan. Usage shows the organization’s credits and activity; billing notices explain limits and required payment actions.
0.9.1
Mobile layouts and investigation fixes
Section titled “Mobile layouts and investigation fixes”- Logs and simulations fit mobile screens better; trajectory fitting and run-event ordering are corrected.
- Meta Agent shows evidence and errors more clearly. Observations and Signals show when Meta Agent is not connected.
- Starting a simulation now asks for confirmation, and stalled queued runs show a warning.
0.9.0
Browser and voice simulations
Section titled “Browser and voice simulations”Create browser or voice scenarios and run them manually or on a schedule. Runs retain the definition used at launch, with a timeline, transcripts and links to application traces. Participants can request sign-in assistance using selected secrets.
Inspect the run evidence to assess the expected result: participants finishing does not by itself establish that the application behaved correctly.
Trajectory View
Section titled “Trajectory View”Trajectory View brings related agent, model and tool steps together across traces, Logs groups and simulation runs. Select a step to inspect its evidence; inferred connections are distinguished from recorded relationships.
Other changes
Section titled “Other changes”- Copy Meta Agent responses and expand grouped activity and evidence.
- Inspect recorded span links in trace details.
- Review organization plans, credits and usage breakdowns separately from recorded model costs.
- Project deletion is blocked while simulation execution or cleanup is unconfirmed.
0.8.0
Meta Agent investigations
Section titled “Meta Agent investigations”Meta Agent investigates project activity and returns answers with references to the evidence. Start from a trace, span, agent or group, ask follow-up questions, inspect tool calls, approve a requested action or stop a turn.
Connect external MCP tools through OAuth or a bearer credential, with permissions controlling which actions need approval.
Observations and Signals
Section titled “Observations and Signals”Observations review recorded activity against your team’s Signals. Define the behavior to check, activate a Signal, and inspect findings with their verdicts and supporting evidence. Deactivation preserves the Signal’s history.
Choose review scope by agent run, trace or a shared attribute in project settings. Run review checks for new evidence without waiting for the automatic review cycle.
MCP access and grouped activity
Section titled “MCP access and grouped activity”External MCP clients can inspect traces and groups, read span evidence, and discover agent runs. Additional credential capabilities enable Signal management and conversation-history access.
Logs group summaries now include child activity in their errors, duration and usage totals, matching the details opened from the group.
Breaking for MCP clients: trace discovery and inspection use cursors instead of offsets. Follow next_cursor and retain the original query bounds when requesting another page; see the MCP reference.
0.6.0
OpenTelemetry tracing for agent applications
Section titled “OpenTelemetry tracing for agent applications”Radiant Platform receives OpenTelemetry traces and brings live execution, agent runs, tool calls and recorded usage into one application.
Copyright © 2026, Radiant AI Limited.