llama-index-tools-ejentum 0.2.0

LlamaIndex tool spec for the Ejentum Reasoning Harness. Wraps the hosted Ejentum MCP server and exposes eight cognitive harness tools as LlamaIndex FunctionTool objects: four dynamic (reasoning, code, anti-deception, memory) plus four adaptive variants (adaptive-reasoning, adaptive-code, adaptive-anti-deception, adaptive-memory) that pre-fit the cognitive operation to the caller's task via an adapter LLM.

llama-index-tools-ejentum

LlamaIndex tool spec that subclasses McpToolSpec and points at the hosted Ejentum MCP server. Exposes the eight cognitive-harness tools as LlamaIndex FunctionTool objects ready for any LlamaIndex agent or query engine.

Use the harness before the agent generates on complex, multi-step, or multi-constraint tasks where the model's default reasoning template would miss a constraint, take a shortcut, or drift across turns. Each call returns a cognitive operation: a structured procedure (numbered steps with a failure pattern to refuse and a falsification test) paired with an executable reasoning topology (a DAG of those steps with decision gates, parallel branches, bounded loops, and meta-cognitive exit nodes). The agent reads both layers before producing its response.

Four dynamic tools (reasoning, code, anti-deception, memory) are available on all tiers including the 30-day free trial. Four adaptive tools (adaptive-reasoning, adaptive-code, adaptive-anti-deception, adaptive-memory) additionally run an adapter LLM that rewrites the matched operation with task-specific identifiers; they require the Go or Super tier.

Tool names exposed to the LLM are whatever the upstream MCP server advertises (canonical hyphenated strings: reasoning, code, anti-deception, memory, adaptive-reasoning, adaptive-code, adaptive-anti-deception, adaptive-memory). This shim does not rename them.

Install

pip install llama-index-tools-ejentum

Configuration

export EJENTUM_API_KEY="ej_..."

Or pass api_key= to EjentumToolSpec(...). Get a key at ejentum.com/pricing.

Usage

Minimal

from llama_index.tools.ejentum import EjentumToolSpec

spec = EjentumToolSpec()
tools = spec.to_tool_list()

Subset of modes

spec = EjentumToolSpec(modes=["reasoning", "code", "adaptive-reasoning", "adaptive-code"])
tools = spec.to_tool_list()

Valid mode names (use canonical hyphenated form): reasoning, code, anti-deception, memory, adaptive-reasoning, adaptive-code, adaptive-anti-deception, adaptive-memory.

With a ReActAgent

from llama_index.core.agent import ReActAgent
from llama_index.llms.openai import OpenAI
from llama_index.tools.ejentum import EjentumToolSpec

tools = EjentumToolSpec().to_tool_list()
agent = ReActAgent.from_tools(tools, llm=OpenAI(model="gpt-4o-mini"))

response = await agent.achat(
    "Why might our microservice return 503s only under specific load patterns?"
)

Tool inventory

Dynamic (all tiers)

Tool	Library size
reasoning	311
code	128
anti-deception	139
memory	101

Adaptive (Go or Super tier)

Tool
adaptive-reasoning
adaptive-code
adaptive-anti-deception
adaptive-memory

Each tool takes a single query: str argument. Returns the injection as a string.

API reference

EjentumToolSpec(
    api_key: str | None = None,
    modes: list[str] | None = None,
    api_url: str = "https://api.ejentum.com/mcp",
    timeout: int = 30,
)

Field	Default	Description
api_key	None	If unset, read from EJENTUM_API_KEY. Raises ValueError at construction if neither is set.
modes	None	Optional subset of harness modes to expose. Defaults to all eight.
api_url	https://api.ejentum.com/mcp	Override for self-hosted MCP gateway.
timeout	30	HTTP timeout in seconds for the underlying MCP client.

The class is a thin subclass of llama_index.tools.mcp.McpToolSpec, pre-configured with the hosted Ejentum endpoint and Bearer authentication.

Wire contract

This shim talks to the MCP endpoint (/mcp), not the direct-REST endpoint (/harness/). For the direct-REST contract used by every other Ejentum shim, see the ejentum-mcp README; for the MCP-over-streamable-HTTP contract, see the MCP specification.

Field structure of an injection and a canonical dynamic-vs-adaptive comparison on the same query are documented in the ejentum-mcp README.

The underlying MCP server

The same MCP server is available on three additional surfaces:

• Stdio via npx -y ejentum-mcp
• Hosted Streamable HTTP at https://api.ejentum.com/mcp
• Listed on the Official MCP Registry as io.github.ejentum/ejentum-mcp

Compatibility

• Python 3.10+
• llama-index-core>=0.13.0,<0.15
• llama-index-tools-mcp>=0.4.1,<0.5

License

MIT

Measured effects

The Ejentum harness is benchmarked publicly under CC BY 4.0 at github.com/ejentum/benchmarks:

• ELEPHANT sycophancy: 5.8% composite on GPT-4o (40 real Reddit scenarios)
• LiveCodeBench Hard: 85.7% to 100% on Claude Opus (28 competitive programming tasks)
• Memory retention: 50% fewer stale facts served (20-turn implicit state changes)
• Plus per-harness numbers across BBH/CausalBench/MuSR, ARC-AGI-3, SciCode, and perception tasks

Methodology, scenarios, run scripts, and raw outputs are all in-repo.