Jev Harness (Apixly)
OfficialServer Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TYPESAFE_API_KEY | No | Optional: only needed for explicitly requested billable run_task inference. initialize, tools/list and this workspace direct route need no key. | |
| TYPESAFE_API_KEY_FILE | No | Optional path to a caller-provided TypeSafe/Jev key file for billable inference; not required for discovery. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| argumentsD | Inspect current argument references for a tool, resolving dependent candidates after supplied parent references. |
| observeC | Read the task state and currently available tools and argument references. |
| operateC | Execute one currently available tool using its supplied argument references. |
| run_taskA | Jev Harness executes the caller's registered task definition end to end for this goal: "Inspect the caller-owned synthetic workspace and save source evidence". The Jev agent loop selects registered tools and fresh argument references from observations, then runs the same independent completion check as direct operation. All inputs and source permissions are already bound by the caller; this call does not need per-step references. One client call returns terminal status, artifact receipt, bounded source quotes when the task defines an evidence report, and actual or unknown model usage. Limits: at most 20 tool effects and 120 seconds, checked cooperatively; callback I/O is bounded by the task. Only available before any operate call. Partial/review/uncertain outcomes are retained, and effects are not automatically retried. |
| read_reportA | Read up to four original quoted records from the current saved evidence report, with source references, character windows and SHA-256. Requires an already saved report that matches this task's records. Available after either route; this does not execute task tools or inference. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 5 tools
observe and arguments both surface task state and argument references, so an agent may be unsure which to call first. operate (single tool execution) versus run_task (full end-to-end loop) also overlap conceptually, though the description of run_task's exclusivity ('only available before any operate call') helps.
Two tools follow verb_noun (run_task, read_report) while observe and operate are bare verbs, and 'arguments' is a noun rather than an action. The set is readable but mixes conventions without a clear rule.
Five tools is well-scoped for an agent harness: one entry point (run_task), a manual loop (observe/arguments/operate), and one post-hoc reader (read_report). Each tool has a clear role in the workflow.
The surface covers the full manual loop plus the automatic path and evidence retrieval, which is strong lifecycle coverage. Minor gaps exist: no explicit cancel/status tool for long-running run_task executions and no way to enumerate registered task definitions.