Data and AI releases

548 this month · 176 companies

Prime Intellect Release notes and new features

Private · San Francisco, US · 36 releases · 1 this month

Website RSS

Also onAcquisitions·Funding(Funding rounds)·Paydays(Pay schedule)

Prime Intellect releases

November 2026 digest

prime 0.20.0

The prime CLI and SDK update includes support for sending Datatrove configurations as pretrain runs, cursor pagination for sandbox lists, and new sandbox egress header rules.

Update
Source ↗

prime 0.19.0

This release adds localhost forwarding to the CLI and SDK, introduces pause and resume functionality for sandboxes, and improves credential file security. It also includes updates to prime-traces and prime-runs.

Update
Source ↗

prime 0.18.0

Prime Intellect released prime 0.18.0, adding storage quota surface controls in CLI and SDK, reasoning effort support and streaming, Slurm cluster resizing capabilities, local training config validation, and sandbox egress policy limit increases.

Update
Source ↗

prime 0.17.0

- Document internet LAN rules in the sandbox CLI and SDK ENG-6615: RWO sandbox volumes SDK and CLI release: cli 0.17.0 and prime-sandboxes 0.5.0

Update
Source ↗

prime 0.16.0

- Add ARM architecture selection to image and sandbox CLI fix(sandboxes): wait for a LAN network's rules when joining, drop joins of deleted sandboxes feat(sandboxes): create a LAN network with its rules feature(train): add prime train resume for paused dedicated runs release: cli 0.16.0 and prime-sandboxes 0.4.7

Update
Source ↗

prime 0.15.0

- render prefixed training logs as readable text feature(train): add prime train pause and --resume-from for dedicated runs feature(sandboxes): join a LAN network at sandbox create time feature(sandboxes): join the create-time LAN network in wait_for_creation fix(sandboxes): stop execute_command queueing past the gateway stream cap release: cli 0.15.0 and...

Update
Source ↗

prime

- improvement(tunnel): read frpc's output with one reader from launch fix(sandboxes): preserve UTF-8 across command output chunks release: prime-sandboxes 0.4.4 feature(volumes): add ls, du, and cat storage commands Don't promise a warm session in volumes create (ENG-6612) [ENG-6372] improvement(train): drop the shared LoRA path and the [deployment]...

Update

prime 0.11.0

- feature(volumes): show who created a volume in list (ENG-6619) feature: surface on-demand capacity and its price in the CLI feat(prime-cli): add prime factory status, workloads, and nodes commands release: cli 0.11.0

Update
Source ↗

prime 0.10.0

- feature(volumes): expand replaces resize, 5Ti default, volume states in list, deprecate --cluster (ENG-6588) feat(volumes): direct R2 get/put, ssh stage progress, warm create (ENG-6585) release: cli 0.10.0

Update
Source ↗

prime 0.9.0

- feature(cli): prime volumes ssh session commands , ENG-6354 feature(cli): prime volumes get / put over the session (ENG-6354) improvement(cli): create a missing --volume when starting a full fine-tune (ENG-6354) feature(cli): reach volume sessions through the public gateway (ENG-6414) feature: prime train accepts SFT configs feature(cli): prime volumes...

Update
Source ↗

prime 0.8.0

- Drop the verifiers dependency and the Lab surface Restore prime deployments and prime fork release: cli 0.8.0

Update
Source ↗

prime 0.7.0

- ENG-5966: Add hosted evaluation lifecycle methods to EvalsClient ENG-5966: release: prime-evals 0.2.4 fix(prime-runs): close failed and cancelled evaluations through the status API fix(cli): default sandbox create image fix(prime-runs): upload exclusively to Prime Traces for enabled accounts fix(cli): verify the upgrade actually changed the installed...

Update
Source ↗

prime-agent 0.9.0

- Fixed background (unattributed) kernel output missing from the expanded IPython cell view: it is now surfaced in the tool details and rendered under a "background output (unattributed)" label after stdout/stderr/result.

Update
Source ↗

prime-agent 0.8.0

- Fixed an OAuth login that finishes after its server was retargeted arming the old-endpoint token against the new URL: credentials are endpoint-bound at issuance, and the host and kernel only use a token bound to the configured endpoint. Breaking: generic MCP OAuth credentials stored before this release lack the binding and require one /mcp login .

Update
Source ↗

prime-agent

Breaking Changes Changed rlm(...) to return at task admission instead of waiting for the child to finish. It now yields a spawn handle (rlm_child_id, name, session_dir, model); RLMResult and its final answer, usage, and model-fallback warning are gone.

Update

prime-agent 0.5.0

Breaking Changes Reworked session input scheduling into a single session action lifecycle and store (daemon protocol 7, schema revision 8); older clients and daemons are rejected cleanly at connect. Changed Changed large daemon session loads to stream JSONL history and avoid retaining a second full-file copy in memory.

Update
Source ↗

prime-agent 0.4.0

Breaking Changes Replaced the recursive daemon get_session_tree response with flat nodes linked by parentId (protocol 6); clients must support the new response shape. Removed /resume and bare --resume; browse sessions with left-arrow from a daemon chat, or use --resume for a direct resume.

Update
Source ↗

prime-agent 0.3.0

- Changed daemon and headless execution to isolate each root session tree in a recoverable worker process, with protocol-v2 chunked snapshots, compact streaming, attachment-local backpressure, session leases, and unchanged print, JSON, and RPC interfaces.

Update
Source ↗

prime-agent 0.2.0

- Improve diffs and traces: show edit diffs in the collapsed IPython view, and make trace uploads observable and resilient with logging, endpoint reporting, and retry on transient failures. Support images in context: paste images straight into the prompt, with validation, resizing, and graceful handling on non-vision models.

Update
Source ↗

prime-agent 0.1.0

- Add a daemon-backed Agents View as the default local entrypoint, with daemon mode and CLI controls for creating, listing, attaching, killing, renaming, and prompting live sessions. Add an AgentConnection client boundary with in-process and daemon adapters, plus versioned daemon protocol metadata and attach snapshots for future Swarm wrapping.

Update
Source ↗

prime-agent 0.0.10

- Add Claude Opus 4.8 to Prime Inference, with reasoning support. Adopt the upstream IPython control-environment prompt: prefer %%bash cells and run project imports and tests through the project's own environment. Rework Ctrl+C to interrupt first and exit on a second press, with Escape now clearing the input bar.

Update
Source ↗

prime-agent

- Fix the onboarding flow so model selection, manual key entry, and cancellation all resolve correctly instead of re-prompting on every launch. Add a Prime Inference onboarding splash for Prime CLI users and persist an onboarding-completed flag.

Update

prime-agent 0.0.7

- Add Python-backed skills that install into the persistent IPython kernel and appear alongside markdown skills. Add Prime team selection during Prime Inference login so team inference costs use the selected Prime CLI context. Guide unauthenticated users through login and model selection before the first agent turn.

Update
Source ↗

prime-agent

- Redesign onboarding: guide unauthenticated users through login and model selection before the first turn, with a refreshed Prime splash and centered menu surfaces. Add a Prime Inference login dialog with browser sign-in and a manual API-key fallback. Adapt interactive surfaces and the editor background to the detected terminal colors (dark vs light).

Update

prime-agent 0.0.4

- Add a system prompt note listing the pre-installed Python packages (requests, httpx, pandas, numpy, scipy, pydantic, and more). Add a default extra-packages constant and kernel bootstrap that installs those packages, with the prompt referencing the constant instead of a hardcoded list.

Update
Source ↗

prime-agent 0.0.2

- Add a persistent IPython tool backed by a Jupyter kernel, with an auto-bootstrapped uv-managed Python runtime, so variables and imports survive across tool calls. Add the RLM harness and prime-agent-runtime bridge so IPython code can spawn recursive child agent sessions via rlm.run, with subagent tray, detail views, and chat-style child transcripts.

Update
Source ↗

prime-agent 0.0.1

Initial Prime Agent release, forked from pi-mono. Add a persistent IPython tool backed by a Jupyter kernel so Python state survives across tool calls, and make it the default tool set. Add recursive RLM subagents: IPython code can spawn child agent sessions via rlm.run, with a compact tray, full child transcripts, and aggregated usage.

Update
Source ↗

prime 0.5.0

- Feature: Add size limit warning for env tarballs fix envshub naming convention Feature: Add sandbox secret integration Added support for push as .rc or .post timeout buffer Feature: support multiple environment installations with prime env install Fix: pre-commit for local dev Feature: Prime MCP integration ENG-1631 expose ports in sandbox pkg feat: add...

Update
Source ↗

prime 0.4.0

- release basic sandbox package move prime core logic to package source inclusion in build buildfix contd.

Update
Source ↗
About and activity

Provides compute, reinforcement learning and evaluation tools so enterprises can build their own AI agents.

Releases per month