RAG architectures are good at one factor: surfacing semantically related paperwork. That's additionally the place they cease.
A framework referred to as a call context graph addresses that hole by giving brokers structured reminiscence, time-aware reasoning, and specific determination logic. Rippletide, a startup within the Neo4j ecosystem, has constructed one. The important thing functionality: brokers which might be non-regressive, capable of freeze validated sequences of actions and compound on them over time.
“The important thing level you need is non-regressivity: How do you make it possible for, when the agent will generate one thing new, you may compound on the earlier discoveries?” mentioned Yann Bilien, Rippletid’s co-founder and chief scientific officer.
Why RAG doesn’t go far sufficient
Enterprise context is sprawled throughout ERP instruments, logs, databases, vector shops, and coverage paperwork. Generative AI instruments can retrieve from all of it — by way of key phrase search, SQL queries, or full RAG pipelines — however retrieval has a ceiling.
Notably, information retrieved is probably not related to the choice at hand (thus inflicting hallucinations); and, even when brokers do pull the precise information, they usually lack steerage to make choices backed by a robust rationale.
That’s, RAG retrieves paperwork, not determination context. “Everybody begins with RAG: Pull related docs, stuff them within the immediate, let the mannequin determine it out,” mentioned Wyatt Mayham of Northwest AI Consulting.
Whereas that works effective for chatbots, it “breaks instantly” for brokers that have to make choices and take actions, he identified. “The most important factor builders battle with is the hole between retrieval and applicability.”
A retrieved doc doesn’t inform the agent whether or not it nonetheless applies, whether or not it’s been outmoded, or whether or not there’s a conflicting rule that takes precedence, Mayham mentioned. “Brokers want determination context, not simply data.”
In development (the human world), which may imply understanding {that a} pricing exception expired, {that a} security coverage solely applies in sure jurisdictions, or that a normal working process was up to date a month prior. “Miss any of that, and the agent confidently does the improper factor,” Mayham mentioned.
With out structured determination context, brokers mix incompatible guidelines, invent constraints to fill gaps, and depend on what Bilien calls "probabilistic guesses over unbounded information." Errors are tough to breed as a result of builders can't hint why the agent made a given selection.
The compounding error drawback is actual, too, Mayham mentioned: A small miss charge per step turns into “catastrophic” throughout a multi-step workflow. “That’s the principle motive most enterprise brokers by no means go away the pilot section.”
How determination context graphs get to the related reply
A call context graph solves this by encoding a structured map of what’s relevant, what the principles are, and once they apply.
The framework is optimized for one query: "Given this example, which context applies proper now?" Time is handled as a first-class dimension; each rule, determination, and exception is scoped to when it’s legitimate.
“The objective is to explicitly handle lacking, incoherent, or contradictory information when constructing the graph to keep away from probabilistic [errors] as soon as the agent is operating,” Bilien mentioned.
The system is constructed round three ideas:
Applicability: Logic is explicitly encoded so the agent is aware of what guidelines to recollect and apply in a given scenario. Context is returned solely when it’s related to the scenario.
Time‑conscious reminiscence: Each rule, determination, and exception is time-scoped. This enables brokers to motive about "What was true then versus what’s true now," then reproduce or clarify its choices.
Resolution paths: The system can clarify the way it obtained from A to B and the "why" behind its rationale (as an illustration, why one piece of context was included and one other was not). Brokers are given "determination path" examples of how comparable instances have been dealt with earlier than.
At setup, unstructured information is ingested and structured into an ontology: what entities exist, what guidelines apply, what counts as an exception. Neuro-symbolic AI handles the sample recognition and encodes formal, machine-readable logic. Over time, the system refines its data base as new choices are made.
“Neuro-symbolic brings two elements: A neuronal half giving a big autonomy to brokers and a symbolic half to cut back the variety of information wanted and convey management,” Bilien mentioned.
The agent is examined at construct time (pre-production) to validate its behaviors or pinpoint enhancements. This reduces dangers in addition to computation wants throughout inferencing, he famous.
Brokers studying, slightly than regressing
In relation to non-regression, the important thing piece is compounding each on intelligence (fashions) and on data (shared between brokers), Bilien mentioned. It’s vital that brokers can discover; once they don’t know tips on how to accomplish a process, they will try completely different prospects, usually in a managed setting or simulation (like a help bot attempting a number of response patterns).
Then, “as soon as an answer is evaluated as passable, the graph freezes that sequence of actions,” Bilien mentioned. Future exploration then begins from this “steady base of validated behaviors” to stop newly-acquired expertise from overwriting beforehand discovered good conduct.
Earlier than an agent acts or impacts a buyer, it checks towards the graph: Is it violating a rule? Hallucinating? Staying inside constraints? Can it generalize the answer throughout comparable instances?
At a macro degree, the system assesses outcomes: Did the conduct enhance long-term efficiency? Did it generalize throughout comparable contexts? Did it protect earlier capabilities?
“This determinism is essential for brokers to run reliability at scale,” Bilien mentioned. It results in conduct that’s extra constant, predictable, explainable, and permitting for stronger management and auditability.
“You need your brokers to have the ability to be taught by themselves once they face one thing they don't know,” he mentioned. “You need them to have the ability to discover and discover new options.”
Getting past "episodic" reminiscence
Whereas the workforce initially assumed it will deploy RL in all places, "that really proved very tough in an enterprise setting," Bilien mentioned. "Knowledge are scarce for some particular use instances and messy for others."
Sometimes, utilizing uncooked information for dependable predictions has been a guide and time-consuming problem, however “now with brokers we entered a brand new period the place constructing ontologies is feasible routinely,” Bilien mentioned.
Traditional supervised fine-tuning strategies can result in oscillations, when fashions neglect the final ability they discovered whereas studying the subsequent tone. General, studying isn’t compounded, compression is “dramatic,” and fashions enhance “episodically” slightly than constantly, main them to repeatedly fail on new or unseen duties.
As Bilien famous: “You’ll by no means have a totally self-learning mannequin if you’re regressing each time.”
In enterprise use instances — like banking the place hundreds of thousands of transactions are processed a day — a excessive degree of reliability is crucial, he famous. “One query I ask all prospects: Is 95% sufficient? In a variety of use instances, it's not. You want 99.999%. 1% off is means an excessive amount of.”
Resolution context graphs can shut that hole, he contends: When the identical buyer help query is requested repeatedly, the agent will return a “passable” reply predictably and with out regression, all whereas retaining autonomy.
Encoding applicability and temporal validity right into a structured graph — slightly than counting on an LLM to deduce it — is a "sound strategy" to an actual limitation in present retrieval frameworks, Mayham mentioned. The open query is whether or not the automated ontology technology holds up towards the messy, numerous information that enterprises even have. "That's at all times the arduous half," he mentioned.
