oceanbase/powercontext

Context for work that humans and agents hand off and continue.

English · 中文 · 日本語

Work rarely ends with whoever starts it. You hand a task to an agent, the agent gets part of the way, and later you or someone else takes over. The reasoning and current state often stay behind in that conversation.

PowerContext keeps context with the work across conversations. When you return, you can see what happened and continue from the current state. A new agent can do the same.

Website · Installation walkthrough

PowerContext 1.3.0 includes the guided setup. The commands below install this version and connect the matching Agent integration.

You see the context the work needs now: confirmed decisions, constraints, progress, evidence, and next steps. You can continue from there or hand the work to another person or agent without rereading the full history.

You decide what will matter later and what needs to move with the task. PowerContext stores durable information as Memory and organizes the current objective and state into a Handoff. You can record reusable approaches as Experience or Skill. PowerContext keeps every item within the scope of the work and preserves its sources and earlier revisions.

The optional recall-sufficiency gate records scope-local daily search effort and budget omissions for later policy evaluation. See the configuration reference for its recording and failure behavior.

You need Git, uv, and your Agent's CLI. Python 3.11+ is required; uv can provision it. macOS and Linux are supported; Windows support is experimental.

Install 1.3.0 and open the interactive configuration wizard in a dedicated directory:

uv tool install --force "powercontext[cli,server]==1.3.0"
mkdir -p powercontext-config
cd powercontext-config
powercontext config init --language en --output .env

If Python dependency downloads are slow or fail, follow the mirror retry instructions.

The wizard asks about storage, local or remote access, memory capabilities, Dashboard, model APIs, and Agent connections. Choose Full memory capabilities to test automatic Memory and Topic Memory; this requires separate Generation and Embedding API credentials. An Agent subscription does not provide those Server credentials. Choose Basic memory to save and retrieve memories explicitly without additional model APIs.

The wizard writes one .env environment file and .env.next-steps.md. If seekdb needs installing, it asks once and installs the dependency in the background. Follow the printed connection details and start Server in this terminal:

powercontext server run --env-file .env

Keep Server running. In another terminal, return to powercontext-config, load only the client settings, and check the connection:

set -a
. ./.env
set +a
powercontext ready
powercontext capabilities

Continue with .env.next-steps.md to create and bind the selected Scopes, install the matching plugins, and launch a new Agent session. The complete walkthrough covers Codex and Claude Code, Dashboard login, SSH forwarding, HTTPS prerequisites, and observable acceptance checks. For example, the matching Codex installation is:

powercontext setup codex --ref powercontext-v1.3.0
powercontext doctor codex

doctor verifies integration setup. To verify automatic memory, check that a real prompt becomes a Source, produces a Topic, evolves after a related prompt, and can be recalled in a new session using the same Scope.

For a Server on another machine, use HTTPS or follow the remote connection guide. Setup recognizes remote HTTP URLs from flags or environment variables and asks for explicit consent; automated setup uses --allow-insecure-http.

Codex is official; other hosts and Python Agent frameworks are community; Bub is evaluation only. These tags describe PowerContext integration maintenance and use. See the capability matrix for supported features and availability.

Codex	Claude Code	MiniMax Code	DeepSeek Harness	ZCode	Kimi Code	Hermes Agent	Pi Coding Agent
OpenClaw	OpenCode	WorkBuddy	Bub	Pydantic AI	LangChain	LangGraph

Applications can use PowerContext through the async Python client, HTTP API, MCP, or the in-process Core SDK. See the interface reference to choose an entry point.

Explore the 22 Chinese Jupyter tutorials and a complete team workflow to run Memory, context preparation, Handoff, Experience, Skill, and a real Agent step by step. The first seven tutorials need no model or API key.

See the methods, full results, and limitations behind these comparisons.

LoCoMo Plus scores with and without Jev candidate filtering after retrieval. Each model is used for both Memory extraction and answering; embeddings use qwen3.7-text-embedding (1024 dimensions). Jev score differences are +8.925, +8.577, and +4.198 percentage points, respectively.

These six scores were supplied by the project owner; the run artifacts are not publicly available for verification. The reproducibility instructions describe the evaluation harness, not verification of these scores.

make install
make check
make test

See CONTRIBUTING.md for the complete development workflow. Coding and memory evaluations, performance benchmarks, and Skill regressions live under evaluation/.

• Get started
• Connect Agents
• Manage context
• Deploy and operate
• Develop with APIs

PowerContext is the successor to PowerMem.

Thank you to everyone who contributes to PowerContext. ❤️

See all contributors · Start contributing

PowerContext is licensed under the Apache License 2.0.