Trajectory
Trajectory is a San Francisco-based platform company that enables continual learning for production agentic AI by converting real user telemetry into post-training data. It serves AI-native enterprises across legal, GTM, customer service, and workforce verticals via an enterprise SaaS subscription and Python SDK.
- Company typePrivate
- Founded2025
- HeadquartersSan Francisco, United States
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What Trajectory does
Trajectory is a San Francisco-based research and product company that has built a platform for continual learning of large-scale agentic AI models. Founded in 2025, the company targets AI-native companies and enterprises that want their production AI products to improve continuously from real user behavior rather than remaining frozen at deployment. Its core product consists of a Web Platform (Instrument, Understand, Steer, Learn) and a Python SDK (pip install trajectory-sdk) that ingest agent traces and product telemetry from sources such as LangSmith, OpenAI, Anthropic, and Vercel AI SDK, standardize them into a "trajectory" primitive, and feed them into training, evaluation, and deployment pipelines. The underlying stack is model-agnostic, supports concurrent multi-LoRA training (C-LoRA) co-developed with UC Berkeley Sky Lab and Anyscale, and incorporates a proprietary extension of Self-Distillation Policy Optimization (SDPO++) for off-policy continual learning.
On the commercial side, Trajectory sells an enterprise SaaS platform on a quote-based annual subscription, with no public pricing or self-service tier. Go-to-market combines direct enterprise field sales (a 30-minute "Book a Demo" motion focused on production AI deployments) with an API-first product surface that lets engineering teams integrate the SDK directly into their products. Named flagship customers include Clay (GTM), Harvey (Legal AI), Decagon (Enterprise Customer Service AI), Mercor (AI Workforce), and Rogo (Professional Services), and the company has secured SOC 2 Type II certification to support enterprise procurement. In May 2026 Trajectory closed a $15M seed round led by Conviction with participation from Bessemer Venture Partners, BoxGroup, Exceptional Capital, and Radical Ventures, alongside personal backing from AI researchers Fei-Fei Li and Jeff Dean.
Trajectory firmographics
Firmographics- Name
- Trajectory
- Legal name
- Trajectory Technologies, Inc.
- Website
- https://trajectory.ai
- Company type
- Private
- Founded year
- 2025
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Trajectory is a San Francisco-based platform company that enables continual learning for production agentic AI by converting real user telemetry into post-training data. It serves AI-native enterprises across legal, GTM, customer service, and workforce verticals via an enterprise SaaS subscription and Python SDK.
- Ownership category
- akta.pro rank
Trajectory industry classification
Industry- Product category
- AI Post-Training Infrastructure
- NAICS
- Software Publishers (5132), Computer Systems Design and Related Services (54151)
- SIC
- Services-Prepackaged Software (7372)
- akta.pro primary industry
- Model Development & Training Platforms (AutoML, Notebooks, Feature Stores) (HDAEANAB)
- akta.pro secondary industries
- Experiment Tracking, Metadata & Model Registry (HDAAABAC), RLHF, Human Feedback & Evaluation Data Tools (HDAAALAJ)
Keywords
Where Trajectory is headquartered
LocationHeadquarters
- HQ city
- San Francisco
- HQ country
- United States
- HQ region
- North America
Offices1 record
Markets served
Trajectory business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Operations
Revenue model
- Platform Subscription: SaaS platform subscription for continual learning infrastructure. Customers pay for access to the Trajectory platform, SDK, and training capabilities. Pricing appears to be enterprise quote-based with no public pricing page.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Subscription | Annual | Enterprise platform access |
Go-to-market motion2 records
Distribution channels1 record
Marketing channels4 records
Trajectory product offering
Product offeringCore offering
Trajectory is a software platform that enables continual learning for AI-powered products by converting real product usage signals—such as user edits, retries, and acceptances—into training data. The offering consists of a Python SDK that ingests agent traces and telemetry from observability platforms and a web-based platform that supports continuous post-training, evaluation, and deployment of large-scale agentic AI models. The system is model-agnostic, decoupling the learning layer from any single base model, and is delivered as a SaaS platform with SOC 2 certified data governance.
Product overview
Trajectory is a platform for continual learning that helps companies continuously post-train large-scale agentic AI models on real product usage. The offering consists of a Web Platform for observing, directing, and crafting AI intelligence (featuring Instrument, Understand, Steer, and Learn capabilities), the Trajectory Python SDK for ingesting traces from LangSmith and custom sources into standardized Trajectory format, and a model-agnostic learning infrastructure layer. The platform turns real product usage signals—edits, retries, acceptances, escalations—into training data, enabling models, harnesses, and prompts to improve automatically. SOC 2 Certified.
Differentiator
Problem solved
Functional benefit
Products and services
- Trajectory Platform Web-based platform for continual learning that provides a creative surface for observing, directing, and crafting AI intelligence. Features four capabilities: Instrument, Understand, Steer, and Learn. Designed for AI-native companies and enterprises with production AI deployments.
- Trajectory SDK Python SDK that turns raw agent traces and product telemetry into a standardized Trajectory format for training, evaluation, and continual learning. Supports ingestion from observability platforms (LangSmith), custom data formats (CSV/JSONL/OpenAI/Anthropic/Vercel), PII redaction transforms, and bulk export workflows.
- Continual Learning Infrastructure Model-agnostic learning layer decoupled from any single base model, supporting open-weight models such as NVIDIA Nemotron and Qwen for sovereign deployment inside customer boundaries. Enables training across model weights, harness, and prompts jointly on production data, with research-driven extensions including SDPO++ for off-policy continual learning.
Quantifiable outcome
- 5x improvement on APEX-Agents benchmark over zero-shot baseline
- +4 more outcomes
Companies that use Trajectory
Customer profileNamed customers5 records
Segments4 records
Ideal customer profiles4 records
Trajectory technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration4 records
AI capability5 records
Feature5 records
Trajectory partnerships and signals
Strategic signalPartnerships
Nine partnerships are on record, tiered core and flagship.
- RunloopcoreTrajectory runs production workloads on Runloop's enterprise AI agent infrastructure platform. The partnership enables Trajectory to run over 10,000 burst concurrent Devboxes for training and fine-tuning workloads, with each isolated microVM spinning up in under 500ms and customer-by-customer isolation maintained.
- ClayflagshipAI-native GTM company testing a model that gets smarter over time from their users. Working with Trajectory on continual learning research and implementation. Already seeing examples of the model learning from its mistakes.
- HarveyflagshipLegal AI company building toward agents that can carry evolving legal domain expertise. Partnering with Trajectory on continual learning for legal work, using Harvey's LAB benchmark for evaluation.
- DecagonflagshipEnterprise AI support company focused on model steerability. Collaboration with Trajectory focuses on rigorously measuring steerability in post-trained models and understanding what training techniques actually improve it.
- MercorflagshipAI workforce company incorporating post-training further up in data production lifecycle. Trajectory's platform allows continuous validation of training signal for new datasets as they produce them.
- RogocoreAI-native company partnered with Trajectory as part of the AI-native companies partnership group.
- UC Berkeley Sky LabcoreCollaborated on concurrent multi-LoRA training platform (C-LoRA). All training code is open-sourced in NovaSky-AI/SkyRL repository. Research collaboration on multi-LoRA always-hot training infrastructure.
- AnyscalecoreCollaborated on concurrent multi-LoRA training platform (C-LoRA). Part of the team that built the multi-LoRA always-hot training stack for continual learning workloads.
- NVIDIAcoreTrajectory's platform works with NVIDIA Nemotron models including Nemotron 3 Super and Nemotron 3 Ultra. Post-training on these open-weight models enables frontier-level performance at a fraction of the cost. Platform is model-agnostic and designed to adopt new open models quickly.
Scale indicators8 records
Recent moves6 records
Expansion highlights5 records
Trajectory competitors and assessment
Company assessmentEmerging players
- Arize AI: LLM and ML observability platform focused on tracing, evaluation, and drift detection for production AI. Comparable around production-trace analysis and behavioral evaluation; complementary rather than fully overlapping with Trajectory's training loop.
Direct peers
- Anyscale: AI compute platform built on Ray for distributed training and inference workloads. Anyscale is already a Trajectory partner (co-built C-LoRA) and shares the same buyer (AI infra teams) and overlapping use cases around scalable, multi-tenant training.
- Comet: Experiment tracking, model monitoring, and ML observability platform serving data scientists and ML engineers. Comparable as a tracking-and-training-adjacent tool for ML teams, though narrower in scope than Trajectory's full continual-learning pipeline.
- Weights & Biases: ML experiment tracking, model registry, and training-orchestration platform used by AI teams to manage training runs and artifacts. Directly comparable to Trajectory as an MLOps platform sitting on top of training workflows, with stronger brand and broader feature surface but less specialization in continual post-training from production traces.
- LangSmith: LLM observability, tracing, and evaluation platform from LangChain — Trajectory explicitly integrates with LangSmith for trace ingestion. Adjacent in the agent-development stack with overlap on trace data, though LangSmith focuses on observability rather than continual training.
- Neptune.ai: Experiment tracking and model registry platform for ML teams, with metadata store and collaboration features. Direct overlap with Trajectory's trajectory/metadata layer and model-lifecycle tooling.
Broad incumbents
- Scale AI: Large incumbent in AI training data, RLHF, evaluation, and post-training services for foundation models. Comparable as a provider of data-and-feedback infrastructure for model improvement, though Scale is much broader, vendor-heavy, and works at the pre-training/foundation layer rather than production continual learning.
- Domino Data Lab: Enterprise MLOps platform with model development, training, and deployment tooling aimed at regulated industries. Comparable as an enterprise-grade ML platform though positioned upstream of deployment rather than at the production-signal continual learning layer.
- Labelbox: Data labeling, RLHF, and evaluation platform for AI training data. Comparable on the human-feedback and evaluation side of the training loop, though Labelbox is more annotation-centric and less focused on production-trajectory-driven continual learning.
- Snorkel AI: Data-centric AI platform with programmatic labeling and weak-supervision capabilities aimed at improving model training data. Comparable in the data-for-training layer and adjacent to continual learning, though Snorkel's specialty is labeled dataset construction rather than production-signal continual learning.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
Trajectory social profiles
Digital presenceTrajectory compliance and trust
Trust signalCompliance1 record
Trajectory financial estimates
Financial estimateRevenue estimate
Valuation estimate
Trajectory leadership team
Management profileNumber of profiles
Profiles3 records
Trajectory funding detail
Funding detailFunding overview
Funding rounds2 records
Investors7 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Trajectory M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Trajectory
What does Trajectory do?
Trajectory is a software platform that enables continual learning for AI-powered products by converting real product usage signals—such as user edits, retries, and acceptances—into training data. The offering consists of a Python SDK that ingests agent traces and telemetry from observability platforms and a web-based platform that supports continuous post-training, evaluation, and deployment of large-scale agentic AI models. The system is model-agnostic, decoupling the learning layer from any single base model, and is delivered as a SaaS platform with SOC 2 certified data governance.
Is Trajectory a public or private company?
Trajectory is a private company. It is classified as venture growth investor backed and is currently operating.
When was Trajectory founded?
Trajectory was founded in 2025. It employs 11 to 50 people.
Where is Trajectory based?
Trajectory is headquartered in San Francisco, United States, in the North America region.
How does Trajectory make money?
One revenue line is on record: platform Subscription.
Who are Trajectory's main competitors?
Arize AI is listed as an emerging player. Direct peers are Anyscale, Comet, Weights & Biases, LangSmith and Neptune.ai. Broad incumbents are Scale AI, Domino Data Lab, Labelbox and Snorkel AI.
Does Trajectory have an API?
Yes. Trajectory API for uploading trajectories, pushing telemetry events, and managing continual learning workflows. The SDK (pip install trajectory-sdk) provides session management (tj.init), trace import (tj.list_conversations, tj.import_conversations, tj.transform), trajectory building (tj.build_trajectory_from_messages, tj.build_trajectory_from_parsed, tj.build_reward_from_scalar), trace workflow (tj.start_trace, tj.upload_trace), and upload/push operations (tj.upload, tj.push_events, tj.save). Supports bulk export from LangSmith, telemetry event streaming, PII redaction transforms, and model ID helpers. Authentication via TRAJECTORY_API_KEY environment variable or trajectory_api_key parameter. Developer documentation is at docs.trajectory.ai/sdk/api-reference.
What industry is Trajectory in?
Trajectory's product category is AI Post-Training Infrastructure. Its primary akta.pro industry code is HDAEANAB, Model Development & Training Platforms (AutoML, Notebooks, Feature Stores), with a secondary code of HDAAABAC, Experiment Tracking, Metadata & Model Registry. Its NAICS code is 5132 and its SIC code is 7372.