Diagnostics

The Diagnostics page — health checks with an overall verdict, the 24-hour model-cost rollup, external MCP server status, interrupted tasks you can resume, and the notifications inbox.

Diagnostics (/diagnostics) is the page to open when something feels off. It answers "is the runtime healthy?" in one place: a set of health checks with an overall verdict, what the agent layer has spent in the last day, whether your external MCP tool servers are reachable, which interrupted tasks can be resumed, and the notifications Forven has raised.

It is an operator page. The dashboard tells you that something needs attention; Diagnostics tells you what.

Forven is a research tool. Diagnostics describe the health of your local install, not the quality of any strategy, and nothing here is financial advice.

Health checks

The top of the page shows an Overall verdict and each check as pass, warn, or fail, with a short message. Click a row for its detail. The page auto-refreshes every 60 seconds; re-run runs every check immediately.

Check	What it verifies
`database`	The SQLite store is reachable and answering.
`auth_providers`	The LLM providers you connected still have usable credentials.
`scheduler`	The scheduler's heartbeat is fresh — jobs are actually being ticked.
`resumable_tasks`	Whether any tasks were interrupted and can be resumed (below).
`recent_costs`	Model spend over the last 24 hours (below).
`tool_truncations`	Recent agent tool calls whose output had to be truncated.
`brain_fence_strips` · `brain_cache_hit_rate`	Whether the brain's prompt caching is working — a low hit rate means it is paying full price for context it has already sent.
`forven_home`	Your `FORVEN_HOME` exists and is writable.

A check that errors internally reports warn or fail, never pass — a broken probe cannot hide a problem.

Cost

A tile shows the model cost over the last 24 hours, with the number of tasks and tokens behind it. This is Forven's own estimate from token counts and list prices — see cost controls — so treat it as a guide and your provider's dashboard as the invoice. For spend per agent over 30 days, see the Overview tab on the Agents page.

MCP servers

The external MCP tool servers you have added under Integrations — the tool providers your agents can call — with each one's status. A disabled server is marked as such. Click a row to manage it.

Resumable tasks

When a task is interrupted — a restart mid-run, a tool that never reported back — Forven records a checkpoint instead of losing the work or pretending it finished. Resumable tasks lists the interrupted work that can safely continue. Click resume to pick it up from the checkpoint.

Some tasks are deliberately not resumable: a task interrupted mid-tool, whose side effects cannot be confirmed, must be reconciled first rather than replayed. A task that was blocked on missing data re-runs its data preflight when resumed, and only proceeds if the data is now there.

# The same data over the API
Invoke-RestMethod http://127.0.0.1:8003/api/diagnostics/snapshot
Invoke-RestMethod http://127.0.0.1:8003/api/diagnostics/resumable
Invoke-RestMethod -Method Post http://127.0.0.1:8003/api/diagnostics/resumable/$TaskId/resume

Notifications inbox

The bottom of the page is the notifications inbox: every notification Forven has raised, newest first, with new ones marked, and each one's delivery log. It is the in-app half of the notification system; Discord delivery, if you set it up, is the other.

From the command line

python -m forven doctor runs a similar set of checks from a terminal (add --json for machine-readable output), which is useful when the UI itself will not load. See the CLI reference.

Steps: triage a problem

Open Diagnostics and read the Overall verdict.
Open any failing check for its detail. A failing scheduler check means jobs are not being ticked — see operations. A failing auth_providers check means an agent's provider lost its credentials — reconnect it on the Agents page.
If Resumable tasks lists work, resume what you still want.
Check the cost tile against what you expect for your throughput preset.
Read and acknowledge anything new in the inbox.

What you'll see: on a healthy install, an Overall pass, no resumable tasks, a cost figure in line with your preset, reachable MCP servers, and an inbox with nothing new.

Operations & system controls — the control strip, processes, and the scheduler
Health monitoring — component states and data-stream SLAs
Notifications — how alerts are routed and delivered
Troubleshooting — recovery recipes for common faults

Diagnostics

On this page